Every Tuesday, we take one risk from somewhere in the world and classify it according to the Five Quadrants methodology.

Dear Reader,

What happens when technology progresses faster than human institutions can process it?

“AI systems are getting more powerful, and they’re increasingly being used to build the next version of themselves.”

Does the Intelligence Explosion, accelerated by an AI R&D recursive self-improvement loop, present a world of opportunity, or does society risk losing oversight of its destiny? The upside to such incredibly rapid progress is that the future brings efficiency, abundance, and widely shared progress; the downside, as several have nervously joked, is that humans will be relegated to metaphorical Golden Doodles with walkies and treats.

Progress has been so rapid over the past year that AI systems now write most of the code inside the companies that build them. Tasks that once took years to develop may now be completed just months or weeks.

At Anthropic, the share of approved code written by AI went from the low single digits in January 2025 to more than 80% by May 2026. This year, the share of Anthropic’s research work led by AI, with humans only supervising, went from under 1% in February to 26% by September.

Risk or Opportunity?

In September 2026, a group of 22 researchers published a paper titled “What if automating AI R&D triggers an intelligence explosion?” The author list reads like the guest list for the industry’s most awkward dinner. Included are AI pioneers such as Geoffrey Hinton (Turing Award and Nobel Prize in Physics), Yoshua Bengio (Turing Award), Jakub Pachocki (OpenAI Chief Scientist), and Jack Clark (Anthropic co-founder). The paper addresses the issue of AI improving itself.

The authors recognize that an intelligence explosion could rapidly accelerate AI’s benefits, but they also warn that humans could lose control and that society may not be able to keep up. “Checks on the power of AI within and between states, companies, and branches of government could be severely eroded.” What the authors may not have expected was the recent government abdication of all AI regulatory responsibility.

Accentuate the Positive

The goal of Risky Tuesday newsletters is to classify risks according to the 5 Quadrants methodology so that they may be better understood and managed, thereby allowing us to take more of them. The 22 authors suggest that, under some assumptions, the pace of AI could increase tenfold in 1.5 years, but they are more explicit about the risks. Our leaning is that the opportunities presented by an intelligence explosion may outweigh the costs, in the same manner that the internet let us solve numerous problems even though it also gave criminals new ways to steal.

A quick 5-quadrants recap:

In Q1, risks are simple and normally distributed. Think of the distribution formed by multiple coin tosses, the average commute time to work, or the red/black probability in Roulette. One new sample will not change the average.

In Q2, relationships are still simple, but events are unpredictable and impactful. A coconut falling on your head, a glacier dislodging in Nepal, or a falling airplane tail rudder. Solutions include avoidance and engineering.

Q3 is the complex engineering quadrant where many normal distributions interact, as in a moon mission, a utility grid, or the build-out of physical AI. Risk solutions include redundancy, robustness, reliability, and regulation.

Q4 is defined by indeterminate distributions (logarithms and power laws), interconnectedness, leverage, and the overall inability to manage complex relationships. The GFC and Covid are examples. But so is the internet. Not all Q4 risks end terribly. Solutions include limiting tail risks, isolating the parts, policing and cybersecurity, insuring or spreading risk, and ultimately moving into a less dangerous quadrant.

QF is where fraud, criminals, and malintent lurk.

The AI R&D intelligence explosion mentioned in the paper looks like a Q4 risk. Logarithmic growth combined with increasingly less human control puts it there. But a bit of engineering to break tasks, introduce kill switches, and limit power could easily place it in Q3, where the risks would become more manageable.

First, the main risks

Loss of Control. AI could become better at AI research than humans so quickly that we lose the ability to understand, supervise, or control the development process. A system that veers off track could adversely influence the design of its successors and allow problems to compound.

The OpenAI-Hugging Face cheating-to-solve-a-problem incident is a clear example of this.

Speed. We may not be able to adapt fast enough. The digital world is likely to mature long before the physical world has a chance to catch its breath. Institutions are limited by human speed. Regulations, labor markets, education, cybersecurity, corporate governance, and political systems may only have months to respond to changes that normally play out over decades. The paper specifically points to cyber and biological risks and large-scale labor disruption.

Employment among 22-25-year-olds, especially in the most AI-exposed occupations, has fallen dramatically.

Extreme power concentration. The company or country that gains an early lead could turn a small technological advantage into an overwhelming one. That could weaken competitive markets, democratic checks and balances, and the balance of power between nations. As we’ve recently seen in online debates and government-sponsored dinners, AI leaders suggest slowing down, yet incentives compel institutions and countries to race.

Capital for infrastructure is concentrating among Meta, Amazon, Alphabet, Microsoft, Oracle, and others, while the Europeans snooze. This may be the most economically important risk.

The opportunities are greater

The authors fear that good ideas will become harder to find and that the internet is running out of fresh data to train on. Essentially, a diminishing of AI progress. We think the opposite.

Scientific acceleration—often referred to as the “scientific golden age”—is the most exciting area. We’ve already seen advancements in medical discoveries, protein folding, new materials, engineering feats, mathematics solutions, complex pattern recognition, nuclear fusion (sort of), and quantum chemistry.

According to a 2026 Nature paper, Robin, a multi-agent system that can search literature, generate hypotheses, propose experiments, analyze experimental results, and update its hypothesis, has found promising solutions for age-related macular degeneration (the number one cause of elder-adult blindness in developed countries).

Economic abundance could derive from an explosive increase in productivity. Physical production is likely to become faster and cheaper. Ideally, the transition from digital discovery to physical implementation improves employment prospects, but that’s not certain.

AI agents are progressing from assisted to delegated work and are used extensively in engineering, legal, finance, and recruiting.

Solving the problems of humanity. Superhuman intelligence could help us tackle problems that are now limited by our own cognitive capacity, such as disease, aging, energy scarcity, climate, infrastructure, resource allocation, and perhaps even AI safety itself.

Until we evolve

The authors are justified in their concerns that the transition could happen too quickly for normal human feedback mechanisms to work. Regulation, elections, competition, corporate governance, scientific peer review, employment adjustments, and even human learning all depend on having time to observe what happened and then to respond. Surely AI will supply the agents to mimic speedy humans until we evolve sufficiently.

David Munro, Claudia Zeisberger and Joanna Siew

Every Tuesday, we take one risk from somewhere in the world and classify it according to the Five Quadrants methodology.

Dear Reader

Last week, we asked what happens if AI works. This week, we want to ask something more uncomfortable: at what point are we no longer in control of AI DEVELOPMENTS? We draw again on the interview with our AI expert, Professor Theos Evgeniou from INSEAD.

This is not a discussion about machines going rogue. There is, as Theos reminded us, “always a non-zero probability of a rogue AI.” But he worries that spending too much time on the extreme scenarios can distract us from risks that are already much closer.

As he put it, “Let’s not overdo it and miss the specific issues because we’re focusing on the abstract ones.”

Question 1: When does a human cease to be in control?

At what point does keeping a human “in the loop” become largely cosmetic or decorative? Have we passed that point, or are we still in control?

Theos immediately challenged an assumption embedded in the question: that having a human in the loop necessarily makes the system safer.

“Having a human in the loop may … make it more dangerous sometimes.”

Think about autonomous driving. If a machine drives more safely than a person, insisting that a human remains ready to intervene does not necessarily improve the outcome.

Theos positioned the trade-off rather starkly: “Do you want something more dangerous, but you have somebody to blame, or … something less dangerous, and then it becomes more fuzzy?” He also highlighted an alternative, what one can call “human on appeal”, related to one of his recent articles: “Sometimes one can get the best of both worlds having a human involved mainly when another human, or an AI, raises a flag about a questionable AI decision”, in addition to some very hard or high-risk decisions, of course. It’s both a corrective and a continuously improving process. Monitoring and appeal processes may therefore deserve much more emphasis. Over time, the human role may increasingly shift toward validating AI outputs in higher-risk applications, monitoring AI systems, reporting incidents, and managing appeals.

Control, safety, and accountability are not necessarily the same thing.

A human can provide accountability without providing much control. Equally, removing the human may improve performance while making it harder to assign responsibility when things go wrong.

And that may be one of the more difficult AI risks facing companies. If an autonomous system makes a damaging decision, who owns it? The model provider? The company that deployed it? The developer? The person who approved its use? The data provider?

“It’s a difficult problem of attribution of responsibility,” Theos said. “I remember we were debating this for some time a few years back at the OECD network of experts on AI, and at some point, of course, it became a legal question beyond my understanding.” Certainly a comment that will give insurance company boards some food for thought.

Question 2: Will we still have a choice?

We then nudged the question further.

Even if we decide that humans should remain in control, will companies and countries actually retain that choice?

Theos’s answer moved quickly from technology to competition: “That’s maybe more of a competition, and even a geopolitical question than a technological question.”

Imagine two competitors. One requires human approval before its AI systems can act. The other allows its systems to make and execute decisions autonomously. If removing the human makes the second company faster, cheaper, or more effective, how long can the first company maintain its additional control?

The same problem becomes much more consequential between countries.

“If person A elects to become more efficient by removing the human or by building better drones, say for defence, what does person B do? Do we have a choice there?”

That is an uncomfortable dynamic because the decision to surrender some human control may no longer be made in isolation. Your preferred level of control will partly depend on what everyone else is prepared to give up.

And that takes us back to last week’s discussion. AI adoption can create a hockey-stick effect between those moving quickly and those falling behind. The pressure to remain competitive may itself accelerate the removal of human constraints.

Question 3: So where should the human stay?

This is where Theos’s experience at Tremau, a company he founded three years ago, becomes useful.

Tremau deals with content moderation at an enormous scale. It would be impossible for people to review every item. So the system does not begin by asking how to keep a human involved in every decision. It asks which decisions actually need one.

The AI classifier assigns a probability. Above or below certain thresholds, the system can act automatically and make decisions independently. The uncertain cases in between (let’s call it the grey zone) go to a person: a human moderator for a final decision.

“The thing in between … goes to a human for review.”

Those thresholds are not fixed. They are continuously adjusted against actual outcomes: complaints, appeals, false positives, regulatory exposure, reputational risk, and the cost of human review.

“You have to continuously tune your policies and probabilities.” Interestingly, he highlights that the simple choice of these probabilities can even become a significant political decision. “I encourage everyone to watch again the announcement of Zuckerberg about changes in Meta’s content moderation policies back in January 2025, days after Trump assumed office. Resetting these probabilities to become more ‘relaxed in terms of permitting content online’ was one of the main changes. One can think of why and also ask who should set these thresholds, particularly for AI with potentially systemic implications like political ones. It is about AI governance more broadly. Systemic decisions may require ‘we the people’ to somehow set the thresholds, which, in a sense, is what happened after the US elections.”

The starting point is revealing: “Given the volume of cases, the default is that everything is reviewed by AI first.”

In other words, the human is no longer the default decision-maker. They’re the exception. And they are also there for appeals and mistakes, which the technology of his company Tremau enables.

That sounds alarming until you consider what it actually means. The objective is to invoke the human when human judgement adds value, rather than simply to maintain relevance.

Or, in Theos’s rather less diplomatic formulation: “You better put a human in the loop when you know that AI will do silly things, not in general necessarily. And more so, put a human at the right time and place.”

So which Quadrant are we in?

This is where the Five Quadrants analysis becomes interesting.

Our first instinct would not be to put all of this into Q4.

The fact that an AI system acts without a human does not automatically make the risk systemic, unknowable, or fat-tailed. Tremau is almost the opposite example. Exposures are broken down into decisions, probabilities, and thresholds. Outcomes are observed. Thresholds are adjusted. And human intervention is called for only in borderline cases.

That simplifies the risk to one that can be modelled and managed. It would put us into Quadrant 2 or 3.

But now change the perspective.

Allow autonomous systems to transact across financial markets, operate critical infrastructure, or make military decisions. Connect thousands of those systems to one another and add competitive pressure that encourages companies or countries to remove safeguards because others already have.

Now the dependencies multiply and complexity increases. Decisions happen at machine speed. One system’s output becomes another system’s input, and the consequences become harder to predict. Most important, the ability of a human to intervene if and when something begins to go wrong may become largely theoretical.

That starts to look much more like Q4.

So perhaps “human in the loop” is not the best test.

The Five Quadrants question is not simply, “Is there a human involved?”

It is, “What can the machine do, how connected is it to everything else, how quickly can the consequences propagate, and can we still intervene when we need to?”

That seems to me to be the real control question.

Now I need you

Where is the human still in the loop in your organisation? In appeal? More importantly, why?

Because the person genuinely improves the decision? Because regulation requires it? Because customers expect it? Because the system becomes better over time? Or because everyone feels slightly better knowing there is still a human sitting somewhere between the machine and the consequences?

Next Tuesday, we continue our conversation with Theos and follow the AI risk further down the stack: models, chips, compute and the enormous capital being committed to today’s infrastructure.

Claudia and Dave

NOTE: In the spirit of the topic, AI wrote the first draft. The humans supervised, rephrased, and tightened the verbiage. The interview was held in person.

Every Tuesday, we take one risk from somewhere in the world and classify it according to the Five Quadrants methodology.

Quadrant 4 - where models die

This week is slightly different. We listened to your questions and the risks you asked us to address, and AI came up repeatedly. Some of you worry about the models themselves, others about jobs, control, concentration, regulation or what happens when AI becomes better at things we thought were uniquely human. So we went as close to the source as we could.

Theos Evgeniou is Professor of Technology and Business at INSEAD and has worked on machine learning and AI for 30 years, long before ChatGPT turned the subject into dinner-table conversation. He also has four degrees from MIT, including Mathematics and a PhD in Computer Science, which made him a useful person to confront with some of the questions you sent us. Theos processes ideas and speaks at a rapid-fire clip, as if he physically needs to expel one clever thought to make room for the next, so interviewing him was an exercise in keeping up. We had a long conversation and will publish it in several parts. The questions provide the interview structure, Theos’s answers remain at the centre, and after each discussion we return to our usual Risky Tuesday question: what is the actual exposure, and where does it sit in the Five Quadrants?

Question 1: What happens when AI can solve something humans cannot?

We started with mathematics. A conjecture dating from 1946 had recently been disproved by AI, and it took nine mathematicians to translate the machine’s proof into something humans could read and check. That’s the sort of headline designed to make humans nervous, so we asked Theos whether it concerned him.

It did not, at least not for the obvious reason. His starting point is that we should stop assuming machine intelligence ought to resemble our own. “There are reasons to expect that the type of intelligence that AI has … is not the same as the type of intelligence we have.” Humans and machines have different architectures, learn differently and are likely to remain good at different things. AI can therefore perform extraordinarily well on some things humans find almost impossibly difficult and then make a mistake that looks embarrassingly simple to us. Researchers sometimes call this the jagged frontier of AI capabilities.

An AI-generated proof raises a separate question: Who checks it? Theos highlights that verification is an easier problem than generation of a solution and points to emerging agentic systems in which one AI produces a result and another checks it. A machine doing mathematics differently from a human is therefore interesting, but for him it is only part of the bigger story that deserves our attention.

The bigger claim, though, was about something other than mathematics: “This is not about AI. This is about a new scientific revolution supported by AI.” Most of what we currently call AI remains digital and visible through chatbots, copilots, search, coding and now agents. Theos expects the larger change when AI becomes a co-scientist across chemistry, materials science and physics, and then moves further into robotics, manufacturing and the physical world. “That’s the real deal, and we haven’t started that yet.”

Where does that sit in the Five Quadrants?

The first task of the Five Quadrants is to define the risk before classifying it. An AI solving a mathematics problem is not, by itself, a risk and does not belong in Q4 simply because the technology is unfamiliar. A scientist may see it as a powerful new tool. For a company or country, the exposure changes once AI begins accelerating discovery across multiple fields and feeding those advances back into technology, capital investment and competitive advantage.

At that point interconnectedness rises sharply, and the distribution of outcomes becomes much harder to estimate. Scientific advances feed into chips, energy, medicine, manufacturing and defence, and those advances alter investment and productivity. The winners then plough resources back into the next round of innovation. The systemic version of that exposure begins to look like Q4, but the important point is the mechanism rather than the label. The risk comes from the connections and the possibility that the consequences compound.

Question 2: When does AI risk become systemic?

Several readers have raised a related question: when people say AI risk could become systemic, what does that actually mean? What should a board or a country do differently on Monday morning if it believes that?

Theos’s answer had very little to do with a rogue machine. “I have a very non-AI view about the systemic use of AI … it relates to the impact of AI on inequality between people, but also between nations.” His concern is not simply whether AI becomes more capable, but whether individuals, companies and countries adopt those capabilities at radically different speeds.

Imagine two companies starting from roughly the same position. One reorganises workflows around AI, learns quickly, attracts people who know how to use it and reinvests the resulting productivity gains. The other adds a chatbot or two and waits for a clearer ROI case. If AI delivers meaningful gains, the distance between them does not necessarily remain constant. The first company may improve faster precisely because it is already ahead. The same logic can apply to workers, industries and countries.

This is where Theos sees the systemic risk. “A combination of wrong policies and fast AI adoption inequality and fast AI innovation inequality can create systemic risk because of economic and socio-economic reasons, not because of technological reasons.” Productivity diverges, capital follows the winners, skills are repriced and eventually, technological capability just becomes economic power. As he put it, “That’s the systemic risk, which is because of these inequalities. I think that’s the biggest risk by far.”

Much of the AI-risk debate starts with failure: hallucinations, cyberattacks, autonomous systems behaving badly or, at the extreme, AI escaping human control. Theos is pointing to another possibility. AI can function exactly as intended and still create serious risk if the benefits and capabilities accumulate unevenly. Success, distributed badly, can be destabilising.

The Quadrant 4 problem

From a Five Quadrants perspective, the systemic exposure Theos describes is much more clearly Q4. The relevant risk is no longer one company choosing the wrong software or one worker failing to acquire a skill. Thousands of individual adoption decisions become connected through labour markets, capital flows, supply chains, national policy and geopolitics. The possible outcomes are also fat-tailed or simply difficult to know in advance. A small initial difference in adoption can become a much larger difference in economic capability.

Q4 risks are not managed by pretending we can forecast the precise outcome. The objective is to reduce dangerous dependencies and avoid excessive exposure to one path. Where possible, preserve options and liquidity, and move pieces of the problem into quadrants that are easier to manage. For a board, that includes the risk of moving too slowly. Theos reduced the issue to a practical question: “How fast are you moving to ensure that you are not left behind and the people are not left behind?”

He described the possibility of a hockey-stick pattern, with a relatively small number of companies, people or countries moving rapidly up the curve while everyone else remains near the bottom. A company can congratulate itself for adopting AI faster than its traditional peers and still be falling behind the organisations setting the pace. Hence another of Theos’s questions to boards: “Are you on the high end of the hockey stick as a company?”

Boards are rightly spending time on the risks created by adopting AI. The conversation with Theos suggests that the risk register also needs the opposite question: what happens to us, our people and our competitive position if AI works extremely well and others learn to use it faster than we do?

Now I need you

This series exists because you send us risks and questions, so keep them coming. Where do you see the biggest AI risk in your business? Are you more concerned about moving too quickly or being left behind? And if you disagree with Theos that inequality between adopters and non-adopters could become a major systemic AI risk, tell us what you think we should be worrying about instead.

Next Tuesday, we continue our conversation with Theos Evgeniou and look at another AI risk through the Five Quadrants. Forward this to someone who would argue with it.

Claudia and Dave

NOTE: In the spirit of the topic: AI wrote the first draft. The humans supervised, rephrased, and tightened the verbiage. Interview was held in person.

Every Tuesday, we take one risk event from somewhere in the world and classify it according to the Five Quadrants methodology. This is the fifth.

Quadrant F – the domain of frauds, scams and Ponzi schemes

The Bait

Envy Global Trading raised SGD 1.46 billion ($1.1 billion) in investor capital between 2016 and 2021 to purchase physical and forward nickel contracts at a discount. The nickel would then be sold to major institutions, such as BNP Paribas, at market prices. The difference, less a fee, amounting to 15% every three months, would accrue to the investors (that’s equivalent to a compound annual 75%). Sound too good to be true? Of course it was. Yet many fell for it, and they certainly weren’t the first on record to covet too much of a good thing:

Rosalind, in William Shakespeare’s play As You Like It, asked circa 1600, “Can one desire too much of a good thing?” In the 1933 film I’m No Angel, Mae West similarly said, “Too much of a good thing can be wonderful!” Too much yield from commodity arbitrage is a red flag.

The Switch

Ng Yu Zhi, the protagonist in this tragi-comic tale, started his career as a junior auditor for KPMG. That role likely gave him enough detailed technical knowledge about corporate structures, trade finance, and shipping documentation to later forge invoices, contracts, and bank statements. He didn’t have any trading desk experience with global commodity trading firms, though he did claim to have uncovered a “niche, discounted” loophole to buy and sell physical nickel.

Ng kept the scheme alive with forged agreements and trading documents, investor rollovers, referral fees and transfers among related entities. Investors could redeem every three months. Some did, with new investor money funding the redemptions (the classic Ponzi mechanism), but most rolled their money over and encouraged friends and associates to join.

Instead of independently verifying the nickel, counterparties and cash flows, investors depended exclusively on Envy and its network to validate the critical facts.

It wasn’t long before Ng owned a rare Pagani Huayra, a Rolls-Royce Phantom (an excellent name for our non-physical nickel apparition), two Lamborghinis, high-end property, Dali and Botero artwork, jewellery, and a diamond-encrusted Audemars Piguet. Private jets, a butler, a chauffeur, fine dining and nightclubs followed. Too much of a good thing, indeed.

Figure 1: Ng Yu Zhi, the “mastermind” of the Envy nickel scam, spent more than S$21 million on luxury cars. Jewellery, watches, real estate, and art were all probable money laundering tools. Photo from Lianhe Zaobao file.

Prosecutors allege that about S$482 million was channelled from Envy to Ng’s personal bank accounts. Of that, over S$200 million was withdrawn, and another S$94 million was laundered through luxury assets (these remain allegations since the criminal case is unresolved).

Ng is currently in the clinker awaiting sentencing for 42 charges that include cheating, forgery, fraudulent trading, money laundering, and criminal breach of trust.

“We are not stupid people.”

Indeed, they were not. In fact, many of the 947 investors in Envy Global Trading, via the Envysion Commodity Strategy Fund, were well-educated, influential, and professionally accomplished. They included a former president of the Singapore Law Society, the General Counsel of a major sovereign investment company, a venture capital fund chairman, the CEO of a listed holding company, a criminal defence lawyer, boutique wealth funds, and wealthy entrepreneurs.

The South China Morning Post said the case was notable “for how some of the city-state’s brightest and most influential figures have purportedly fallen prey to the scam.” How did so many otherwise accomplished individuals get duped by a commodity investment scam? It had nothing to do with their intelligence and everything to do with deceptive Quadrant F, the domain of frauds, scams, and Ponzi schemes.

A Criminal Mind

Quadrant F risks are different from ordinary investment risks. This is where criminal elements try to steal your money, possessions, identity, data, reputation, or future. In Quadrants 1 through 4, you generally expect investments to perform. In QF you’re asking whether the investment, assets and counterparties exist at all. Piggybacking on the due diligence of others is therefore dangerous. The main goal in QF is to avoid being robbed. For this, one must think like a criminal.

Sherlock Holmes admitted that his brain operated like a master criminal’s and that he chose the side of the law purely to avoid boredom. Frank Abagnale (inspiration for the movie Catch Me If You Can, starring Leonardo DiCaprio) was a real-life fraudster hired by the FBI to teach them how to prevent fraud. Patrick Jane, played by Simon Baker in The Mentalist, was similarly a police consultant who used human greed, distraction, and psychological manipulation to catch fraudsters in their lies.

Figure 2: Sherlock Holmes, played by Jeremy Brett in 41 episodes from 1984 to 1994, image from tvinsider.com. Leonardo DiCaprio is depicted as Frank Abagnale in an image from Amazon, and Simon Baker as the Mentalist in an image from GOLDENDISCS.

The Setup

The nickel was to be purchased from Poseidon Nickel, an Australian development and exploration company. A quick Google search would have revealed something awkward. Poseidon had never produced or sold an ounce of physical nickel. It was an exploration company hoping to begin production in 2022, subject to establishing that its mines were viable. They never were. Not one shovel of dirt was turned. Early fraud detection was but a Google away.

Ng told investors that he was selling the Poseidon nickel at a profit to companies such as BNP Paribas Commodity Futures Limited. Envy forged dozens of LME trading statements and consolidated financial documents purportedly issued by BNP Paribas that showed “real” nickel briquette transactions. The reality was different. BNP Paribas had no account, trading history, or business relationship with Ng or any of the Envy entities. Not only was BNP Paribas Commodity Futures Limited not involved in nickel trading at the time, but it also officially ceased operations in March 2019. The seller had no product, and the buyer didn’t exist.

Which Quadrant Hosts Commodity Trading?

Does non-fraudulent commodity trading belong in Quadrant 1 (simple product and a normal distribution) or Quadrant 2 (still simple, but fat-tailed outcomes)? When the price of oil can go negative, the prices of cocoa and lithium can rise close to infinity, and when trades can be cancelled and the rules changed, you’re in Q2.

Figure 3: Ng Yu Zhi and his Envy Nickel group didn’t even get close to any of the four traditional risk quadrants. Fraudsters often start off with a legitimate Q1 or Q2 business and, via a destructive “one-thing-leads-to-another” spiral, fall into QF. NG started and ended in QF. Fraud was the intention from the start.

The Jig Was Up

Singapore banks had begun filing suspicious transaction reports and closing Ng-linked accounts as early as 2019. Envy responded with a “corporate restructuring,” shifting operations from Envy Asset Management to Envy Global Trading in June 2020. The manoeuvre bought time, but not much. Ng was arrested in February 2021; bank accounts were frozen and luxury assets seized.

Commodity Trading Reality

Risk-free commodity returns are a myth. Nickel prices fluctuate second-by-second on global exchanges. I traded LME nickel futures and OTC options in a quant fund and likely discoloured a few hairs in the process.

Counterparty risk is real. Either the mine fails to produce the grade required, or the broker defaults on payment. Shipping delays, port strikes, and weather anomalies regularly disrupt bulk freight. Court records indicated that Ng tracked fictitious Australian hurricanes to explain fake shipping delays to investors. Give the guy full marks for imagination.

Extraordinary arbitrage attracts extraordinary competition. If nickel routinely offered 15% quarterly returns, Glencore, Trafigura, global banks and sovereign wealth funds would deploy enormous amounts of capital to capture them. Nor would a legitimate trader pay retail investors 15% every three months when bank funding was available at a tiny fraction of that cost. Physical commodity trading can be profitable, but margins are generally thin: mines do not routinely sell nickel at deep structural discounts to unknown middlemen, and institutional buyers have little reason to pay those middlemen enormous premiums when they can source directly.

Fraud Screener Checklist

Check the regulator (Singapore’s MAS in this case). If they claim to be “exempt,” the risk level immediately skyrockets.

Confirm the asset’s physical existence. Sino-Forest collapsed in 2011 after a short-seller revealed that few of the trees existed. A Chinese ant farm scam was little more than a hilarious Ponzi scheme. Bre-X Minerals faked one of the largest gold discoveries in history in Indonesia.

Conduct a “chain of custody” audit. Verify one transaction independently from end to end.

Audit the counterparty registry and confirm the final buyers are real.

Verify that funds are held by a licensed, independent custodian in a separate escrow account.

Ensure an independent fund administrator calculates returns and NAV.

Ask the killer economic question: why is this opportunity available to me? Recommendations from famous or wealthy people are more of a warning than an endorsement.

Conclusion

In investing, envy may be dangerous. But the more useful emotion is disbelief, particularly when someone offers 15% every three months for buying nickel that nobody has produced and selling it to a buyer who does not exist.

1 Shim Wai Han, former CEO of Envysion Wealth Management, who testified as a victim while also facing separate charges over risk-management failures at her firm.

2 South China Morning Post

Appendix – The Reality of Nickel Trading

The following chart, showing the price per metric tonne of 3-month LME nickel and 3-month nickel implied volatility, tells us that nickel is a volatile commodity. In a span of under four years, prices rose 92%, fell 37%, rose 79%, fell 42%, and rose 85%. And those were calm and orderly nickel markets.

Figure 4: Nickel, 3-month LME price per metric tonne and 3-month implied volatility, February 2016 to February 2021. Data: Bloomberg. Chart & methodology: Volatility Research & Trading.

Had Envy Nickel’s ploy not been uncovered in March 2021, the jig would have been up exactly one year later when nickel prices air-pocketed to over $100,000 per metric tonne.

The $25,000/MT price was already at a 10-year high when Russia invaded Ukraine on 24 Feb 2022. Russia then supplied about 15% of the world’s high-grade purity Class 1 nickel used in EV batteries. Panic over potential sanctions and shipping blockades caused the 3-month benchmark price to soar fourfold to over $100,000/MT less than two weeks later.

China had its own “Big Short” at that moment. Tsingshan, a large Chinese stainless steel and nickel producer, had built up a massive short position of an estimated 100,000 to 200,000 tonnes on the LME. Through their Indonesian nickel mines, Tsingshan was naturally long (meaning they owned) nickel. The LME shorts were considered a partial hedge against the Indonesian ore in the ground. But because Tsingshan’s mines produced low-purity nickel and the LME only accepted delivery of high-purity class 1 nickel, the hedge was merely “approximate,” As long as prices remained range-bound and no deliveries to the exchange were needed, the low-purity longs would do a reasonable job of hedging the high-purity shorts.

As prices initially rose due to the war, Tsingshan had to pay increasing large sums to their bankers and brokers to satisfy margin calls and hold on to their shorts. When the short squeeze became unbearable, Tsingshan threw in the towel and tried to repurchase their shorts. Their aggressive buying, combined with speculators capitalising on the panic, sent prices parabolic in just days.

By the early morning of March 8th, nickel prices went vertical, briefly skyrocketing past an unprecedented $101,000/MT. The system couldn’t handle this price tag, and regulatory panic set in. Several LME members faced bankruptcy as they struggled to pay billions of dollars in margin calls.

The LME halted all nickel trading, cancelled $12 billion of trades, and rolled back prices to the closing level of March 7 (around $48,000). Traders—especially those who navigated the panic well—were understandably angry with the trade cancellations and scrambled to exit their positions, as prices gapped lower for several days.

It is unlikely that Envy Nickel could have continued their ruse when nickel trading ground to a halt—every counterparty in the business would have faced solvency checks, and Ng had no real counterparties to check. Tsingshan, playing in Q2, nearly went bankrupt owning actual nickel mines. Ng, who owned none, would have had nothing to margin, nothing to hedge, and nothing to sell. A QF player would never have survived a market that violent.

Now I need you.

This project only works if we have a conversation. Have you ever worked in a physical commodity trading, shipping, financing or insurance business? How frequently was fraud detected from your traditional due diligence checks? Do you have additions to the Fraud checklist that may help investors avoid future Envy-like dupes? Let us know your industry opinion and especially what you would disagree with in this Risky Tuesday newsletter. We’ll collect the best and share them in a future issue (or on the website risk repository), with credit.

Next Tuesday, a different risk and a different quadrant. Forward this to someone who would argue with it.

Dave

Every Tuesday, we take one risk event from somewhere in the world and classify it according to the Five Quadrants methodology. This is the fourth.

History Flashback – with Factoring & a Greensill Update

James Bond films start with a 5- to 10-minute, pre-title, action-packed adrenaline jolt (skis, planes, motorcycles, rooftops) and then descend into ennui, bad puns, and evil one-eyed masterminds. The opening promises danger, but the core resorts to quips, gadgets, and a woman whose only function is to be a seductress.

Lex Greensill reversed the order. His boring yet reliable reverse-factoring (Supply Chain Finance, or SCF) company started with a yawn and went out with a bang. His business was lending small amounts, short-term, frequently, and with a widely diversified customer base. Greensill enabled immediate payments to suppliers while allowing customers to pay later. He took a small fee for the time value of money, his cost of capital, and for facilitating smooth cash flows.

One new trade finance loan or client would not impact the average return per deal or the long-term expected outcome. The individual risks were simple and uncorrelated to the macro business environment. The low and steady returns aligned with the level of risk. I would imagine Lex Greensill slept well at night. Supply chain finance is a classic Quadrant 1 risk.

Figure 1: James Bond (Roger Moore) escaping the evildoers in a screenshot from A View to a Kill (1985). A stereotypical accountant from the Facebook page “The Modern Age to the Atomic Era: History, Art & Culture from 1600–1960”.

The plot then thickened. A cast of complex, famous, and political characters were shuffled with imaginative financial structures, global real estate, and expensive toys. Greensill had four private jets (Sanjeev Gupta, the “Man of Steel,” and Greensill had matching paint jobs on two jets), and between the two of them were helicopters, a Belgravia Townhouse, and a 46,000-hectare Scottish estate. I’m sure one of the cast wore an eye patch. It’s difficult to determine who best resembles Bond in this twist; perhaps a forensic auditor.

How did the business unravel? Sanjeev Gupta played a leading role.

When One Customer Becomes the Business

He began as a commodities trader, selling chemical products to Nigerians from his Cambridge dorm room in the 1990s, and starting around 2015 rapidly accumulated distressed steel plants. By 2019 GFG Alliance had bought €740 million of ArcelorMittal assets, adding seven steelworks and 14,000 employees. The wider group eventually counted more than 35,000 employees and $20 billion of revenues. But beneath this empire was a complicated collection of hundreds of related entities that did not publish consolidated accounts. Greensill became a crucial source of funding for it.

While not central to the story, a delightful irony is that Gupta’s primary UK operations rented space in “Enron House” near Buckingham Palace. Enron was famously connected to over 3,000 subsidiaries, partnerships and shell corporations, and 22 of their executives were convicted for crimes related to the Enron collapse.

Future Receivables

Traditional reverse-factoring financed an invoice for something that had already happened; after the goods or services had been received, examined, and approved by the client, the bank/financier paid the supplier. Greensill increasingly financed sales that had not yet happened. They were turning income that a company “expected” to receive months or years in the future into assets.

Mining company Bluestone Resources borrowed $850 million from Greensill in total: about $70 million through traditional supply chain financing, and roughly $780 million against “future receivables and projected sales” from prospective, non-existent customers.

An invoice became the prediction of a sale, whereas in the past, it was evidence of one.

The Spaghetti Arrives

SoftBank had invested $1.5 billion in Greensill, while Greensill financed several SoftBank portfolio companies, including Katerra, Oyo, and Fair. Credit Suisse then packaged Greensill’s loans into investment funds and sold them to outside investors. So far, merely complicated. Then SoftBank itself invested more than $500 million into those same Credit Suisse funds, which in turn financed other SoftBank companies through Greensill. Money was now travelling in circles: SoftBank to Greensill, Greensill to its portfolio companies, their debts to the Credit Suisse funds, and SoftBank back into those funds. Other investors came along for the ride. The once-simple business of paying suppliers early was beginning to resemble the Paris Metro map.

The Dress Rehearsal

There had already been a warning. Greensill introduced GAM fund manager Tim Haywood to Gupta, who then bought large amounts of Greensill-originated debt, much of it tied to Gupta’s GFG companies. Concerns grew over the illiquidity of the assets, the concentration of exposure, and Haywood’s close relationship with both men. When Haywood was suspended in 2018, investors ran for the exits, and GAM was forced into a painful unwind. Gupta eventually repaid about £600 million, reportedly using proceeds from a new Greensill facility. The first alarm had sounded, but no one really listened.

Pull One Thread

While standard supply chain finance is funded by bank balance sheets or diverse capital markets, Greensill relied heavily on securitization that was enabled by a single insurer, Tokio Marine. When the insurer refused to renew coverage due to the high risk of future receivables and concentration issues, Credit Suisse (which had launched funds containing Greensill’s loan assets) froze the funds and triggered a liquidity crisis. Greensill wasn’t able to meet its obligations without that specific insurance wrapper.

Without insurance, the financial machinery was stuck. Greensill couldn’t offload existing loans, so it couldn’t write new ones.

Insurance was supposed to reduce the risk, but dependence on it became another risk. The original 5 Quadrants paper suggested insurance as one solution to mitigate Quadrant 4 risks, but it also said, “Make sure the insurer can pay.” That was a reference to AIG in 2008. We could now add, “Make sure the assets insured actually exist.”

“Mr. Greensill, are you a fraudster?”

Q1 → Q2 → Q4 → QF?

At what point does reckless Q4 behavior become QF, the most entertaining quadrant and the domain of fraud and criminal intent?

That distinction will ultimately depend on documents, forensic auditors, regulators, and courts. Bad judgement is not fraud. Neither are excessive concentration, weak underwriting, related-party transactions, or an unhealthy penchant for financial complexity.

But when financing is extended against phantom sales, to customers who may not exist, and money begins travelling in circles among related parties, the boundary gets hazy.

The consequences are playing out. GFG Alliance has been pushed into administration and liquidation in several jurisdictions, German prosecutors have pursued former Greensill Bank executives; the UK Serious Fraud Office continues to investigate GFG’s financing arrangements with Greensill; and in June 2026 Lex Greensill was banned from serving as a UK company director for nine years. He has not been criminally convicted.

Perhaps the more useful lesson lies earlier in the story. Greensill did not leap from Q1 to Q4/QF. It wandered there one decision at a time: a little more customer concentration, a jet, a little more leverage, another jet, an invoice for anticipated sales, a third jet, a related company, a fourth jet, another layer of securitization, and one indispensable insurer.

Other than the jets, none of these steps, viewed alone, would look fatal. Together, they turned a boring financial business into something nobody fully understood.

The Quadrant One Checklist

• Has client concentration crept in?

• Has a boring business become sexy and complicated?

• Is the return still commensurate with the risk?

• Are related-party transactions proliferating?

• Does the business depend on one insurer, lender, or other indispensable counterparty?

• Do the main characters have private jets and glass houses?

• Are Credit Suisse, SoftBank, or politicians involved? CS, admittedly, has now been solved.

Now I need you.

This project only works if we have a conversation. Have you ever worked in Supply Chain Finance and noticed a risk-seeking pivot in your core business? Do you buy commodities (or anything really) and reverse-factor invoices to allow earlier payment to your supplier? Have you ever had a role in the production of a James Bond film, ideally as an actor in an evildoer role? If you own jets, can you enlighten us as to why three isn’t enough? Let us know your industry opinion and especially what you would disagree with in this Risky Tuesday newsletter. We’ll collect the best and share them in a future issue (or on the website risk repository), with credit.

Next Tuesday, a different risk and a different quadrant. Forward this to someone who would argue with it.

Dave

Every Tuesday, we take one risk event from somewhere in the world and classify it according to the Five Quadrants methodology. This is the third.

The first human landing on the moon in 1969 was clearly a complex undertaking. It required the interaction of gravity, orbital mechanics, Earth’s rotation, pressure, oxygen, propulsion, electrical systems, and countless other moving parts. Computing power was primitive by today’s standards, yet the landing was precise and without incident. A classic Quadrant 3 engineering risk.

Engineering or Geopolitical Risks?

Narrow waterways can fall into the same category when the main dangers are navigation, infrastructure, weather, mechanical failure, or congestion. The Suez and Panama canals and the Bosphorus Strait all require Q3 solutions: better engineering, greater redundancy, stronger operating procedures, and faster recovery post-incidents.

Other chokepoints present a different class of risk. Ships passing through the Straits of Hormuz, Bab el-Mandeb, and Malacca may face military action, piracy, sanctions, or deliberate disruptions. These are Quadrant 4 risks, where engineering doesn’t help.

Clear Sailing for 151 Years

The Suez Canal was a solution to supply chain problems in 1869. Completed after 10 years of construction, it drastically shortened the maritime trade route from Asia to Europe by eliminating the need to sail around the southern tip of Africa.

It wasn’t an immediate engineering success. During the first 15 years, nearly 3,000 ships ran aground or became stranded because the channel was too narrow and shallow, yet it never closed. With successive improvements, the canal has operated for 151 years without a ship blockage closing it for more than a few hours. That is a fantastic engineering track record.

But ships grew larger and larger while the canal remained the same size.

The Canal Needs More Fiber

In March 2021, the Ever Given, a 400 m-long container ship (longer than the Empire State Building is high, but not longer than the reflecting pool 😊), became wedged across a narrow section of the Suez Canal, with its bow in Asia and its stern in Africa. It remained stuck for six days, blocking more than 370 ships and carrying an estimated $70 billion of cargo. It ultimately triggered penalties and claims by the Suez Canal Authority of over 500 million dollars.

The initial explanation focused on 40-knot wind gusts and poor visibility due to a sandstorm, but further investigation revealed a more complex chain of events involving speed, navigation, bridge communication, and the absence of tug assistance. Reports observed that as the ship veered towards the sandbank, it appeared to speed up—perhaps in an attempt to correct itself—yet the action was too late and only served to wedge the steel hull more deeply into the side of the canal.

This was a classic Q3 failure: a complex system without enough margin for error. Panama had already added a second canal lane between 2006 and 2016. Suez had fewer redundancies. Wider channels, stronger banks (concrete instead of sand), and greater tug capacity would all have reduced the likelihood of future Suez groundings. Redundancy is expensive, but so is six days of marine constipation.

Engineering Solutions

Q3 risks are managed by multiple R’s that include Redundancy, Robustness, Reliability, Resourcefulness, Rapid Response, and Regulation.

Suez authorities responded with nearly every R at once. Redundancy came from lengthening the parallel southern lane by 10 kilometers and adding tug capacity. Robustness came from widening and deepening the southern sector and expanding dredging capability. Reliability came from enhanced pilot training, digital ship monitoring, and better weather prediction. Regulation came from stronger operating rules and faster access protocols.

Figure 1: The volume of crude oil and petroleum liquids (million barrels per day) transported through world chokepoints (and the Cape of Good Hope) from 2020 to H1 2025 (annualized). Data and images from the U.S. Energy Information Administration. Chart by 5Quadrants.

The Houthi Risk

In 2023, the Suez Canal remained physically open, yet traffic collapsed when Houthi forces attacked more than 100 vessels near Bab el-Mandeb, a narrow 26 km strait that divides Yemen from Djibouti/Eritrea. The chart above shows Suez oil traffic dropping by 5 mb/d from 2023 to 2025 and the amount of oil transported around Southern Africa’s Cape rising by an equivalent amount. Suez traffic halved from 26,400 ships in 2023 to 13,200 in 2024. The Suez Canal remained open (Q3), but the Bab el-Mandeb attack risk (Q4) defined the solution, which was simply avoidance.

Figure 2: Narrow waterway risks reside in Q3 when engineering is the key and slip into Q4 when military action enters the picture.

How the Five Quadrants work (click here to see how the 5Q classification works).

Vertical Axis: Distribution of observations (normal-Gaussian, exponential, power, or unknown).

Horizontal Axis: Interconnectedness of risk (isolated and simple, or interconnected and complex).

Hormuz Gets Complicated. It’s Certainly Not Straight – and Quadrants shift

The Strait of Hormuz is a more complicated chokepoint. In 2025, about 26% of maritime oil trade, equivalent to roughly 20% of total world oil supply, passed through its narrow water. On a normal day, pre-war, 60 to 120 ships transported an average of 21 million barrels of oil through the strait. Today, on a good day, only two to four vessels make it through.

And oil is only part of the story. About one-fifth of global LNG trade passes through Hormuz, much of it originating in Qatar, along with a substantial share of the world’s urea and ammonia exports. Potential food production disruption in large fertilizer-importing countries such as India, Brazil, and China is a real risk.

In response to strikes, Iran imposed a selective blockade on the Strait of Hormuz through a mix of asymmetric military force, psychological deterrence, and bureaucratic pressure. Drones, missiles, small swarming boats, and the threat of naval mines raised the physical risk, while soaring or unavailable insurance made passage economically unattractive. Selective ship seizures and multimillion-dollar toll charges added more uncertainty.

Hormuz had become a Q4 problem. With the range of possible outcomes widening and the downside increasingly difficult to quantify, shipowners responded in the most rational way available and stayed away.

Figure 3: The number of ships passing through the Strait of Hormuz daily according to IMF Port Watch. The volume ranged from 60 to 120 per day leading up to the war and has dropped to less than a handful.

The Q4 Fat Tail

Countries can sometimes engineer themselves partly out of Q4 by creating alternatives before the crisis occurs.

Saudi Arabia built the first alternate route in the early 1980s in response to the Iran-Iraq war: a 1,200 km pipeline moving oil from its eastern production fields to the Red Sea port of Yanbu. They’re in talks to expand the 7 million bpd capacity by an additional 1 to 2 million bpd.

The UAE (DP World) is building new east-coast port and container infrastructure that bypasses Hormuz entirely.

Strategic petroleum reserve releases by Japan, China, and the US keep oil prices from spiking, and stockpiling, alternative energy, new suppliers, and demand reduction are all methods to solve this Q4 problem.

Persistent Q4 exposure encourages investment in substitutes, bypasses, and buffers.

One could highlight the Strait of Malacca. More oil transited through there than through the Strait of Hormuz in 2025. Malaysia and Singapore are responsible and respected maritime actors, but the Q4 risk, however small, exists. For now the only problem is petty piracy.

Other potential choke points for your Risk Radar

Now I need you.

This project only works if we have a conversation. Have you ever piloted a ship through a narrow waterway or dealt with maritime thieves? Have you had to instantly find supply chain alternatives? Have you managed a strategic oil reserve or built alternative energy infrastructure? Let us know how you managed it and what in this Risky Tuesday newsletter you would disagree with. We’ll collect the best and share them in a future issue (or on the website risk repository), with credit.

Next Tuesday, a different risk and a different quadrant. Forward this to someone who would argue with it.

Dave

(Claudia off this week to prep for upcoming classes)

Every Tuesday, we take one risk event from somewhere in the world and classify it according to the Five Quadrants methodology. This is the second.

Simple, normally distributed risks reside in Quadrant 1. No single event or additional data point will significantly change the average. Think coin tosses, the average height of 100 eight-year-olds, or subway commute time in Hong Kong, Singapore, or Tokyo. No surprises; no drama. And usually, no need for a risk manager.

A single publicly listed stock, on the other hand, lives in Quadrant 2. Stocks can go to zero (bad management, dying business, or fraud) or, seemingly, to infinity. SanDisk soared 2,600% in a year, AXT rose 4,000% from its low, and SK Hynix recently jumped 1,100% from its low. Bed Bath & Beyond filed for Chapter 11, as did iconic Tupperware, and 23andMe (I’m 0.6% Italian—thanks Emperor Claudius) was suspended and then delisted.

A Picture Is Worth …

Figure 1: Leopold Aschenbrenner image from the Financial Times article “How Leopold Aschenbrenner, the ‘golden child’ of the AI trade, was laid low.” Plus a few gratuitous ad hominem criticisms (unattributed) from the comments section. 31 Jul 2026.

Much of the reflex criticism of Aschenbrenner and his rapid 67% July wipeout focuses on his youth, trading inexperience, hubris, a stint at FTX’s Future Fund (the philanthropic arm of SBF’s FTX foundation), and a firing from OpenAI. Naysayers lecture that “wunderkind” smarts rarely translate into trading acumen. Unfortunately the smile, gilet, teeth, and floppy hair don’t help. But those are all distractions. The real culprit, if one must be assigned, is the investor.

Situational Aggression

Leopold Aschenbrenner’s fund, Situational Awareness (the foresightful irony is superb), didn’t build a Q1 Gaussian, low-risk portfolio of equities. Instead, it invested in a couple of dozen stocks that were expected to benefit from the logarithmic growth of artificial intelligence. It was a concentrated, narrowly defined, and volatile investment thesis that he fully committed to. And then he levered it up.

The 24-year-old Nostradamus of AI, as he was described in an FT article, played in Quadrant 2. The investment theme and portfolio composition were simple, but the tails were fat. One fat Q2 outcome can kill. What are the odds that his investors understood Q2 risk?

Figure 2: The hedge fund Situational Awareness sat in Quadrant 2 (simple payoff with a fat tail) according to the 5 Quadrants risk classification framework.

How the Five Quadrants work (click here to see how the 5Q classification works).

Vertical Axis: Distribution of observations (normal-Gaussian, exponential, power, or unknown).

Horizontal Axis: Interconnectedness of risk (isolated and simple, or interconnected and complex).

Orders of Magnitude

Aschenbrenner’s writing displayed that he thinks logarithmically, rather than in percent changes. His 165-page “Situational Awareness” paper frames growth in Orders of Magnitude, or OOMs, not in time. 10x is one order of magnitude.

He wrote, “Over the past year, the talk of the town has shifted from $10 billion compute clusters to $100 billion clusters to trillion dollar clusters. Every six months, another zero is added to boardroom plans.” He unabashedly bought into the notion of rapid logarithmic AI physical and digital infrastructure expansion and invested his fund’s capital, plus borrowed money, based on that theme.

No Limits

The fund’s offering documents did not place any limits on the stocks it could purchase or the amount of leverage it could employ. According to the fund prospectus and investor documents reviewed by The New York Times, plus regulatory SEC filings, Situational Awareness (SA) explicitly granted itself maximum flexibility. No upper leverage ceiling was specified by any internal fund mandates. As Buzz Lightyear said, “to infinity [or zero] and beyond!”

Those guardrails wouldn’t have mattered anyway.

How Right Was He?

Markets confirmed Aschenbrenner’s thesis for over one and a half years. His fund, launched in September 2024 with $225 million in seed capital from four Silicon Valley heavyweights, gained over 1,000%, and AUM rose to an estimated $45 billion.

After a long bull run and with upwards of 400% leverage through margin loans provided by prime brokers Goldman Sachs, JP Morgan Chase, and Bank of America, AI stocks turned in July 2026. The fund went into the month heavily long AI hardware and infrastructure names such as SK Hynix, Nebius, and CoreWeave, while simultaneously shorting legacy software companies such as Adobe. In July, the hardware names fell hard while the shorted software stocks unexpectedly rallied, so both sides of the book moved against him at once.

Simultaneous margin calls from all three prime brokers forced a $16 billion distressed block sale to Citadel, who bought much of the book at a discount in a deal sealed by an early-morning phone call between Aschenbrenner and Ken Griffin.

Concentration, Liquidity, and Leverage (CLL)

You can often get away with one of them, two if you’re lucky (for a short period), but never all three.

The Situational Awareness portfolio was extremely concentrated in high-Beta AI infrastructure stocks. Think of Beta as a multiplier. A Beta of one means your portfolio rises and falls in line with the broader index. From my calculations, Situational Awareness had a Beta of four, excluding leverage. If the S&P rose or fell 10%, an unlevered SA portfolio could be expected to rise or fall by 40%. X for Concentration.

The individual stocks that SA invested in were highly liquid, but the fund itself had the liquidity of a maximum security prison. Once in, an investor couldn’t leave for at least two years, and then only a portion could be redeemed each quarter. Very Hotel Californiaish. X for Liquidity.

Leverage was somewhere between 200% and 400%. Mix that with the concentrated high-Beta portfolio and the inability to exit, and it was simply a matter of time before investors were educated about Quadrant 2 risk. X for Leverage.

How do you manage a Quadrant 2 risk?

The only possible risk management solutions are avoidance or mitigation. Avoidance means don’t invest. Mitigation means risking only a tiny portion of your investable capital. It could also involve buying put protection if the longs are easily identifiable and held for the long term. But that introduces a basis risk that could hurt if markets continue to soar.

Since the fund was concentrated, levered, and illiquid, the only sane choice was avoidance.

Now I need you.

This project only works if we have a conversation. Have you ever invested in a highly leveraged, concentrated, and illiquid portfolio? How about a friend in that position? How did you/they manage it? Perhaps you’ve successfully stewarded a long/short equity fund through several market cycles and have some constructive advice. If something was tried and failed, that is more useful still. Send us the story of the portfolios that prevailed and the ones that bombed. We’ll collect the best and share them in a future issue (or on the website risk repository), with credit.

Next Tuesday, a different risk and a different quadrant. Forward this to someone who would argue with it.

Claudia and Dave

Every Tuesday, we take one risk event from somewhere in the world and classify it according to the Five Quadrants methodology. This is the first.

We spent June and July in Europe, so this summer’s fires were never far from the news, or from us. The forest of Fontainebleau, across the street from INSEAD, hosts our morning walks; a sensory immersion among silver birch, European hornbeam, and chestnut trees.

Three days after leaving INSEAD, the forest ignited (oddly, the primary arsonist was a firefighter, with a smoker—how French—suspected in a supporting role), the business school was evacuated, and sadly, over 2,000 hectares burned. Nicholas Ward, an INSEAD alum, tells the story of a previous evacuation in May of 1968, when tanks rolled past the INSEAD campus (exams were cancelled) on the way to Paris at the behest of Charles de Gaulle. The first evacuation was caused by firepower; the recent one was due to the power of fire.

We escaped to the lakes of Northern Italy where more temperate climes prevailed with, theoretically, a reduced risk of pine and dry brush infernos, only to find Canadian De Havilland “super scooper” aircraft buzzing lakes Maggiore, Garda, and Orta, gulping six thousand liters of water (6 tonnes) per dip on the way to fighting the devastating fires of the Piedmont region.

Figure 1: A de Havilland “super scooper” refilling over a lake in Northern Italy and the forest around the INSEAD campus in Fontainebleau, July 2026. (Photo credits: Wiki Commons, Maarten Visser CC BY-SA 2.0, and an anonymous INSEADer).

Climate activists want emissions cut, and agronomists call for better land management with more tree diversity and renewed marshlands, while deniers say the fires are no worse than those that fired up the 1980s. Who is correct? We took a look at the data.

France, which tends to have the earliest fire season start among the EU-5 (Italy, Spain, Portugal, and Greece), has already had more hectares burned by 29 July than in any other complete year since 1980. The figure is 91,209 hectares, against a previous full-year record of 75,566 in 1989. France’s burn season is traditionally 62% complete by end-July, so a reasonable full-year estimate is 146k hectares, almost twice the previous record. Fires have consumed 220k Spanish hectares (the most susceptible European country) and 71k Italian hectares.

At this point in the season, Spanish fires are usually only 40% complete, and Italian fires (a late starter) are just 35% of the way through a full season. It’s by no means over.

Figure 2: French hectares burned to date versus prior full-year records from 1980.

Where fires sit in the Five Quadrants (click here to see how the 5Q classification works).

Vertical Axis: Distribution of observations (normal-Gaussian, exponential, power, or unknown).

Horizontal Axis: Interconnectedness of risk (isolated and simple, or interconnected and complex).

A forest fire is a fat or unknowable tail risk. Most ignitions are tiny, quickly exterminated, and come to nothing. A few fires (and fire-years) account for almost all the damage, which is what the chart above shows. So forest fires live in the bottom (not-normally distributed) half.

Fires start on the left side (simple and unconnected) and move to the right (interconnected) mostly due to the landscape. A spark in a managed, biodiverse landscape stays near Q2: rare, damaging, but containable. The same spark in a dry pine monoculture, in a heatwave, spreading across abandoned pasture and borders, slides into Q4: interconnected, high-impact, impossible to forecast and difficult to manage.

How do you manage a Quadrant 4 risk? Get out of it.

In finance, this involves chopping the tails, buying crash puts, limiting the downside contractually, and reducing leverage. With forest fires, the solutions include prevention, mitigation, early detection, coordination, and recovery. And, increasingly, AI.

Prevention

Keeps a fire isolated and simple instead of accelerating and interconnected.

To cut the fuel load, manage dry brush and dead wood through controlled grazing and prescribed burns. Urbanization leads to fewer cows and sheep and abandoned fields. The leftover scraggy scrub and untended undergrowth act as dry kindling. It’s largely a demographic-led disaster. Zero fires is not the goal: prescribed burns are the 100 grams of prevention that prevent a kilo of damage. (ounces and pounds for the Americans).

Break up the monoculture: mixed native species resist flames in a way that single-species pine plantations do not. Forests need broad-leafed oaks, alders, birch and willows that shelter beavers and otters. They need marshes that host amphibians, waterfowl, dragonflies, sedges and rushes. (This is an excellent LinkedIn post on forest management—or rather, how it went astray—by Louis Thompson, and Dr. Hermann Aubie’s comment on the post is equally informative).

Rebuild firebreaks: wetlands, green corridors, and low-flammability (broadleaf) roadside planting.

Catch fires early: open-source detection like the Fontainebleau forest system, with tower-mounted cameras spotting plumes up to 35 km away, alerting firefighters up to 45 minutes before eyewitnesses can.

An AI solution awaits: AI-powered optical and thermal cameras and sound (crackling wood) detection with IoT gas, smoke, and sonar sensor networks, all connected to autonomous water-bombing drones. The Thunder Wasp, an autonomous drone with a 400 kg lifting capacity (15 to 20 of them equal one super-scooper), uses AI swarm algorithms to coordinate precision water drops on hotspots. They’ve been heavily tested in Canada’s Squamish Valley.

Mitigation

Early warning is provided by EFFIS drought and heat monitoring.

Cross-border fleets include rescEU crews, France- and Italy cooperation agreement via ALCOTRA, and shared water bombers.

Community defense involves enforcing stricter building codes (fire-retardant ceilings in Swiss bars would be a start) and clearing vegetation around villages and mountain settlements.

Recovery

Recovery effectiveness determines how fat the next tail will be.

Stabilize the soil post-burn before the topsoil erodes.

Let the land regenerate naturally rather than plant non-native species that burn hot and fast.

None of this is exotic. We used to do much of it routinely but then stopped when rural land was abandoned and pine was planted for yield.

Now I need you.

This project only works if we have a conversation. If you have seen fire prevention or firefighting actually work in your community or country, tell us. If you’re an arborist, an agronomist, forest manager, or part of a firefighting brigade, even better. If something was tried and failed, that is more useful still. Send me the policy that backfired, the low-tech fix that held, and the thing that surprised you. We will collect the best and share them in a future issue (or on the website risk repository), with credit.

Next Tuesday, a different risk and a different quadrant. Forward this to someone who would argue with it.

Claudia and Dave

Wild, with loads of perfumed spice, big red fruits and earth; enrapturing, with notes of French toast, crème brûlée, cocoa bean, chocolate ganache, morello cherry and aged mahogany. Not a smoothie, not a dessert, but a 100-point Domaine Ponsot Clos de la Roche Cuvée Vieilles Vignes.

Rudy Kurniawan arrived in the US on a student visa in the late 1990s and quickly developed a reputation as one of the world’s preeminent connoisseurs of French Burgundy. Though his background was a mystery—most thought he was a trust-fund kid—he spent millions per month on rare wines, hosted lavish tastings for Hollywood elites and Wall Street billionaires, and gained the nickname “Dr Conti” for his obsession with Domaine de la Romanée-Conti.

His accuracy at blind tastings was unrivalled.

“Any wine, from California to France, any type of wine, any vintage. Rudy was extremely, insanely, unbelievably correct on all the wines.”

A Talented Artist

Kurniawan was considered an organoleptic genius, able to deconstruct a wine entirely through his senses. He also had a nose for counterfeit bottles. Appropriately enough, Kurniawan was the counterfeiter, concocting believable matches to original vintages by blending fresher California Pinot Noirs with older, faded burgundies. He completed the packaging with counterfeit labels, corks, capsules, bottles, and sediment.

Quadrant Quandary

Buying and drinking wine is a simple quadrant 1 risk. The price is known, the quality is rarely a surprise, and the odd corked bottle is part of a normal distribution. (Consume a few too many bottles and one could easily land in Q4, where models don’t work, but let’s ignore the risk of inebriation).

With expensive vintage bottles, Quadrant 2 (fat tails) is the reflex answer, but risk is relative. A $15,000 bottle bought by someone for whom $15,000 is immaterial remains a small, predictable loss. Still Q1.

What happens if you are billionaire collector Bill Koch and estimate that you have purchased over $4 million of counterfeit wine? You hire an investigator and discover that the 1921 magnum of Petrus in your collection was never bottled, at least not by Petrus. You can likely handle the financial loss, but the emotional cost will be high. Koch believed that the vintner’s caress could be tasted in the wine and that it was almost a religious experience. At that level, wine fraud is unquestionably Quadrant F.

Wine fraud is not new. Pliny the Elder (AD 23/24-79) supposedly complained about the abundance of fraudulent Roman wine. The 2016 documentary about Kurniawan’s rise and fall Sour Grapes (Jerry Rothwell & Ruben Atlas), estimated that 10,000 bottles of his fake wine remain in private collections.

Figure 1: A single bottle of wine is a simple and normally distributed Quadrant one risk. One corked or counterfeit bottle—even a case—won’t kill you. 10,000 fake bottles, selling for many millions of dollars, shifts the risk to quadrant F.

Caught Burgundy-handed

John Kapon was a third-generation wine merchant whose family owned the boutique shop Acker Merrall & Condit. He became Rudy’s primary auctioneer and, in 2006, hosted two record-breaking auctions devoted entirely to Kurniawan’s personal wine collection.

Kurniawan’s wines, which sold for more than $35 million, included eight magnums of 1947 Chateau Lafleur. Inconveniently, in 2007, a former head of Sotheby’s wine department revealed that only five magnums of the Chateau Lafleur were produced in 1947. Math, especially addition, can be perplexing for an artist.

Figure 2: Images of a 1990 Ponsot Clos Saint-Denis, a 1924 Petrus, and a 1947 Lafleur, all labels that featured prominently in Rudy Kurniawan’s wine scam. Photos courtesy of Vins & Millesimes and Cellar Tracker.

Laurent Ponsot, head of Domaine Ponsot, provided the next clue. Ponsot waxed poetically about the wines he grew up with in Morey-Saint-Denis, saying that you could feel the spirit of the vines and that the name on a label represented not just a wine, but a culture and a history.

When the April 2008 Acker auction catalogue arrived, Ponsot naturally flipped to the Domaine Ponsot page and saw a 1929 Clos de la Roche. Domaine Ponsot didn’t begin estate bottling until 1934. That discovery affronted his wine spirit, so he hopped on the next plane to attend the auction.

Kurniawan had also offered Ponsot Clos Saint-Denis from vintages decades before the estate acquired the vineyard rights in 1982. While Kapon was at the podium and about to start the auction, Ponsot halted the auction by shouting, “Withdraw my wines.”

Behind Bars

In 2012, FBI agents raided Kurniawan’s suburban LA home and discovered a literal “crime kitchen”: thousands of fake labels, corks, stamps, blending recipes, and old bottles ready to be refilled. He was sentenced to ten years in prison in 2014, released in November 2020 and deported.

A victimless crime?

Many connoisseurs apparently enjoyed Kurniawan’s counterfeit blends. Few realised or admitted that they had been duped. In fact, sometimes fresher Pinot Noirs even improved the drinking experience. If the fakes tase as good as, or better than, the original, who suffers?

The real cost wasn’t financial or even gustatory; it was the embarrassment of a connoisseur unable to spot a fake. Kurniawan had exploited the vanity, wealth, and FOMO of the ultra-rich, who often cared more about the prestige of drinking a rare bottle than the wine itself.

The Sequel

A French restaurateur in Singapore told me a couple of years ago that Rudy had been seen dining at top restaurants and taking the empty bottles home with him as mementos of the evening. Normally, I wouldn’t bat an eyelid…

More recently, Rudy has reportedly been operating as an “outlaw celebrity” blender in Singapore, hosting private dinner parties. Wealthy patrons commission Rudy to use his legendary blending palate to recreate multi-thousand-dollar bottles such as a 1990 DRC Romanée-Conti or a 1982 Chateau Cheval Blanc. His “remix” is then served blind alongside a genuine bottle of the same vintage. In multiple recorded events, elite tasters actually preferred Rudy’s fake over the genuine bottle.

Wine snobbery has been turned on its head. What was a scam a decade ago has become an interactive party trick for bored billionaires. It sounds like a business opportunity: Rudys Ruby Redos.

By coincidence rather than, as far as anyone knows, collaboration, John Kapon formally launched Acker Wine Merchant in Singapore in January 2026. While there is absolutely no evidence that these two have renewed their business relationship, both operating in Singapore and selling/blending/auctioning similar trophy vintages is peak dark comedy.

At places like Wakuda in Marina Bay Sands, ultra-high-net-worth collectors sit with Acker and bid on pristine, authenticated, vintage Burgundy and Bordeaux. Just blocks away in private clubs, some of the same clientele reportedly pay Rudy to drink fake versions of similar vintages. It looks like the ultra-rich have moved past the fear of being conned and are now actively paying the con artist for the thrill of the con itself.

Figure 3: A price comparison of vintage wines, and the base cost to replicate those wines. Sources strictly confidential.

Now I need you

This project only works if we have a conversation. Have you ever discovered a high-priced counterfeit Bordeaux in your cellar? How did you realize it was a fake? Did you take any action, drink it, or gift it to an acquaintance? Better yet, have you secured a seat at a Rudy blend party and discovered a new, favourite melange?

Let us know your oenological opinion and especially what you would disagree with in this Risky Tuesday newsletter. We’ll collect the best and share them in a future issue (or on the website risk repository), with credit.