The noise around AI doomerism increased this week as a post by former OpenAI and Anthropic engineer Jacob Coxon went viral. There was nothing particularly new in the post. It was the same warning about how fast the technology is moving and that it could soon self-develop into a superhuman system capable of ending humanity within the next decade.

Anthropic, the company behind the AI Claude, has been publicly saying similar things for years, and it is spending tremendous resources pushing for AI regulation.

What sparked as much, if not more, controversy was the apparent coordination surrounding the post. Skeptics and conspiracy theorists rightly pointed out several unusual findings:

Was this a well-planned psyop or simply a well-intentioned insider trying to warn humanity of our impending doom? It is becoming increasingly difficult to tell. Pre-COVID, I assume most of us would have given this story the benefit of the doubt. Post-COVID, and in an AI-uncertain world, it is much harder to know.

How can we assess people, truth, and risk in an increasingly uncertain world seemingly optimized to mislead us? Is there a framework we can use to assess such situations?

I’ve studied and used game theory for decades. It’s been invaluable in investing, entrepreneurship, and understanding how people behave when their decisions affect one another.

Game theory isn't a catch-all for strategic thinking or decision-making under uncertainty.

It specifically studies situations involving two or more players where each player’s outcome depends, in part, on what the others choose to do.

For example, game theory wouldn’t apply to solitaire. It absolutely applies to poker.

In poker, you rarely know exactly what another player is holding. Instead, you use what you can observe. This could include their actions, incentives, tendencies, and past behavior to estimate a range of possible hands. From this information, you decide how to respond.

You are constantly assessing whether an opponent has a legitimately strong hand or is bluffing. What do their behavior and actions tell you about the cards they’re holding and how they stack up against everyone else? The strategy is to put the other players on a range of possible hands they have or could potentially have.

For our purposes, let’s simplify the range of possible poker hand strengths into four categories:

Can we use game theory and poker to better understand how we should think about AI doomer claims from supposed insiders and companies like Anthropic? What potential range of “hands” could they be playing?

By hands, I mean: How likely is it that they are on track to produce AGI, or artificial general intelligence, powerful enough to seriously harm humanity? Are their concerns legitimate, or are they overhyping the risk for other reasons? And if we believe the risk is real, how should we respond?

I think we can eliminate the possibility that Anthropic or other AI doomers are completely bluffing. AI technologies are amazing, capable, and improving by the day. The risk of them doing something bad, even if it isn’t catastrophic, is non-zero.

From the severity of their public comments, they clearly want everyone to believe in the monster or premium-hand scenario. I believe individuals like Dario Amodei, the co-founder of Anthropic, actually believe what they’re saying. His company has invested billions of dollars developing a product that he continually warns could cause unprecedented economic disruption or even possibly kill humanity. Why would he do that?

Their stated rationale is similar to the nuclear arms race strategy after World War II. The race to AGI is winner-take-all, and the good guys must win. And, of course, they’ve anointed themselves as the lone good guys—the only ones capable and moral enough to develop it safely. That is the definition of a god complex, in my opinion.

Anthropic maintains a public Responsible Scaling Policy document on their website and they’re lobbying at both the federal and state level to have it adopted. It is hoping to define the rulebook for everyone.

Anthropic is also pushing for strict government-mandated third-party audits that are expensive and favor well-funded incumbents like itself.

It also wants to standardize outcome-based deterrence, which would impose severe legal liabilities for model deployment while shielding itself and others who defer model approval to the state. This directly targets the open-source community, which is one of the biggest threats to Anthropic. Open-source model providers can’t control what others might do with their models, and the liability alone could completely shut down the open-source market.

These proposed actions are right out of the regulatory-capture playbook historically used by big companies hoping to thwart competition.

From this lens, it feels more like they’re playing with a marginal hand, where they recognize real threats to their business model and hope to use fear to build a moat around it. There are plenty of facts to support this theory.

First, it’s becoming increasingly likely that the value generated by AI will ultimately not accrue to model providers like Anthropic. Intelligence itself might end up being the commodity layer within the AI ecosystem. The value is currently flowing to compute—computer chips and data centers—energy, and the harnesses and software that run agents. Anthropic is pushing into the software and harness areas, but it has had to purchase compute from companies such as xAI.

On the model front, open-source models are getting better and keeping pace with frontier models. Not to mention the back-and-forth among frontier competitors like OpenAI. Mere weeks ago, everyone would have said Anthropic had the best models. With the recent release of OpenAI’s GPT-6 Astra, that’s no longer true. As I’ve written about previously, model independence is a standard everyone should be building toward, and many are doing so.

Also, Anthropic’s regulatory efforts are heavily U.S.-focused. Of course, it has called for international limits, but we all know China would never agree to any such deal. If you can’t stop proliferation globally, what is the point of hindering it domestically if your real concern is a winner-take-all scenario for AGI?

What if they’re right? Where should we place our chips if they are?

Historically, centralizing power has not gone well for humanity. If a powerful super AI is inevitable, it needs exposure to the environment in which it is built so that the environment has a chance to develop immunity.

Suggesting government oversight is a joke. Coming from the investing world, I still remember the Securities and Exchange Commission directors who dismissed the Bernie Madoff whistleblower, claiming that “headquarters had Ph.D.s who understood derivatives.” Clearly, they did not. Do we expect government officials to understand AI?

I would place my bets with the thousands of engineers and white-hat hackers constantly pressure-testing these models.

By maintaining a vibrant open-source community, widespread AI access creates a collective "immune system," allowing millions of developers to find and patch vulnerabilities faster than a few corporations can. Giving defenders the same advanced tools as potential bad actors creates a decentralized, resilient defense that eliminates single points of corporate failure.

AI is here, and it’s only progressing faster. Free markets have a considerably better track record of producing safety and prosperity than centralized authoritarian regimes. That’s where I am placing my bets.

My goal with The Leap is to provide you each Saturday with the knowledge, tools and lessons learned to help you get started and keep going toward building your future. 

Whether you are making the leap to startups, solo-entrepreneurship, freelancing, side hustles or other creative ventures, the tools and strategies to succeed in each are similar.