In poker, the most expensive mistakes don't come from bad cards. They come from great stories.
When the guy across the table leans in, voice tight, eyes wide, and tells you with the utmost conviction that he's got you crushed — sometimes he's telling the truth. More often, he wants you to believe that he’s truthful. The tell isn't the words, it’s in how hard he's working to make you feel something before you've had the time to properly analyze the situation yourself.
The same is playing out in AI today.
Last week, AI researcher Jacob Coxon resigned from Anthropic and posted a warning that lit up every feed that matters: frontier labs are racing toward self-improving superintelligence, the people building it earnestly believe it could kill us all by the end of the decade, and continuing the race is gambling with our lives.
The Wall Street Journal ran an exclusive, and many alignment researchers piled on. Evan Hubinger, Anthropic's Alignment Science lead, publicly agreed that people inside the building really do take extinction-level risk seriously.
We should take the substance of AI risk seriously. I do. But the story we’re being told is at best incomplete. After Coxon’s initial warning came the second, more interesting development.
Within 24 hours, and despite only having a few hundred followers prior, Coxon’s post had leapt into nine-figure view counts. Media outlets, politicians, and academics latched onto it immediately. Bernie Sanders and other congresspeople came out with several AI-related bills, all under the guise of improving safety.
If it doesn’t seem organic, it probably isn’t.
With the help of tech guru Brian Chau, online sleuths mapped Coxon to Newspeak House, a London civic-tech organization The Guardian once described as trying to "hack" British politics. It’s an NGO funded in part by the Long-Term Future Fund and the EA Infrastructure Fund, the same effective-altruism money networks that keep showing up next to AI-doom advocacy.
Others flagged the improbable virality of a quiet account, the speed of amplification from safety-policy orgs, and how short Coxon's Anthropic tenure actually was. Former colleagues pushed back that he's a real researcher with real OpenAI experience who actually cares.
I’ve spent years in the AI ecosystem, both as an operator and more recently as an investor, and firmly believe that we should be clear-eyed about both the upside and the failure modes. Systems that can improve themselves, acquire resources, and operate outside the intent of their operators are fundamentally a different category of technology than a better chatbot.
All of those things can be true at once, and none of Coxon’s affiliations necessarily disproves the risk claims. But it does remind us of useful heuristic: any time that tools of power are in play, as is certainly the case with frontier AI, the narrative is also the product.
And products have sellers.
This week the product got a sequel as Anthropic, OpenAI, and Google have begun publicly discussing an industry-led body that would set AI safety standards — talks that reportedly date back to July — while CEO’s Amodei, Altman, and Musk publicly argue we need to "pace the frontier.”
The labs racing hardest are telling everyone the race is too fast, and quietly workshopping who gets to write the rules. Doom spikes attention, which drives visibility, valuation, and influence. A standards body consolidates pace, access, and legitimacy.
When someone reminds you to be terrified or euphoric, consider who benefits from you believing that story.
AI is the greatest opportunity in a generation for organizations to secure money, control, and power. Labs want capital and talent. Policymakers want jurisdiction. Corporate media, academics, and advocacy groups want budgets (and to feel like they’re not languishing on the sidelines). The real fight is over money, control, and who defines "safe enough."
The companies themselves aren't neutral narrators. "We need to win the race so the responsible party gets there first" is a story. "Slow down or we all die" is a story. "This will cure cancer and create abundance" is a story. Every one of them is designed to move your emotions before it moves your analysis, an no lever has proven more effective than superintelligence.
Poker taught me something that has held true during my 20 years in the tech/startup ecosystem: the harder someone pushes you to adopt a specific emotional frame, the more carefully you should evaluate their hand.
So what's the answer? It isn’t "ignore AI risk”, it isn't "trust the labs”, and it isn't "trust the doomers." It’s refusing to outsource your judgment to whoever is best at making you feel something their desired emotion, and instead asking what incentive the narrative is serving before you buy the prophecy.
What specifically is being claimed, and what’s the evidence for it?
Who gains money or control if this framing wins?
What would falsify the claim? If nothing, you're not looking at analysis.
Are concerns being raised regarding today’s systems, or hypothetical future systems doing rhetorical work for the power brokers?
Where is the steelman for the other side, and was it even attempted?
Net net: AI will mint extraordinary fortunes and create extraordinary channels to wield power. There’s real risk, and real upside. To best navigate the coming world, don’t outsource your thinking to the best storyteller.
The road less traveled right now isn't absolute certainty. It's thoughtful consideration.
RECENT DETOURS
Companies — Just backed uranium startup Subatomic. Nuclear power is perfect primed for a comeback. The demand for energy is unprecedented, modern tech has made nuclear safer than ever, it’s the best accelerant away from fossil fuels, and the global stakes for energy independence have never been higher. The team is highly experienced in the operational and regulatory realm, and it’s backed by an incredible team including Joe Lonsdale from Palantir. Email me if you’re interested in learning more.
Sips — 2023 ELLMAN Caryn Renae Sauvignon Blanc. I’m generally not a big Sauvignon Blanc drinker, but this is an exception — and my favorite domestic version none. It’s got body and weight more similar to Chardonnay, but paired with the crisp, refreshing flavor of Sauvignon Blanc. Truly the best of both worlds and perfect for the last rounds of summer heat.
That’s all for this edition.
Cheers,
Greg