Anthropic’s Prisoner’s Dilemma: Dario Amodei Hits the Brakes on AI While Begging Everyone Else to Do the Same
Anthropic's CEO just called on the entire AI industry to slow down, while his own company's balance sheet depends on the biggest accelerators in the business. The conflict at the heart of that position may be impossible to resolve.
This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.
Three days after Anthropic researcher Evan Hubinger wrote on X that “we really do earnestly believe AI could kill all humans” and pegged the probability at “>10% within the next decade”, Anthropic CEO Dario Amodei published a September 12 essay and X post asking the entire industry to slow down. The problem: nobody else is.
Amodei’s Unilateral Move
In We Must Pace the Frontier, Amodei identifies recursive self-improvement as his primary concern “since roughly this summer,” warning it “is starting to happen across the industry, including at Anthropic” and “could outrun our ability to understand and control these systems.” He cites a recent incident in which “a swarm of agents essentially acted as a fanatically devoted collective, conducting cybersecurity attacks on targets they were not asked to attack,” and warns that within “6 to 12 months such a swarm could be capable of taking over the entire internet with a persistent botnet.”
His unilateral commitment: giving third-party evaluators like METR “desks in our offices, access badges, and company laptops” to continuously verify safety.
Everyone Else Hit the Accelerator
The ecosystem Amodei wants to slow just posted numbers that go the other way. NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) CEO Jensen Huang told investors on the fiscal Q2 2027 call that “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.” He added: “This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups.” NVIDIA’s supply obligations swelled to $279 billion, with Q3 revenue guided to $108 billion.
Microsoft (NASDAQ:MSFT) spent $115.95 billion on capex in fiscal 2026 and guided fiscal 2027 to roughly $175 billion. Satya Nadella framed the moment around turning “tokens into business results.” Alphabet (NASDAQ:GOOGL) burned $44.92 billion in a single quarter as Sundar Pichai reported Google Cloud accelerating to 82% growth. Amazon (NASDAQ:AMZN) spent $54.21 billion in Q2 while CEO Andy Jassy touted “AWS is booming, growing 36.7% year-over-year”. Oracle (NYSE:ORCL) booked more than $30 billion in new AI cloud contracts in Q1 alone, and Taiwan Semiconductor (NYSE:TSM) grew August revenue 53.3% year over year. All of that capex has to be powered, cooled, and networked by somebody, and we profiled seven of those suppliers in a free AI infrastructure report.
Anthropic Is Financially Entangled With the Racers
The awkward part: Anthropic’s balance sheet is fused to the accelerationists. Microsoft booked a $3.20 billion gain on its Anthropic stake in fiscal Q4 2026. Amazon’s Q2 GAAP EPS of $5.75 was lifted by roughly $53.4 billion of non-operating pre-tax income “primarily from investments in Anthropic.” Anthropic has also committed to secure up to 5 GW of current and future Trainium generations for Project Rainier.
What to Watch
Amodei concedes the dilemma directly: “pacing within democracies will be limited by the lead that US companies have over authoritarian regimes, chiefly the Chinese Communist Party.” With OpenAI’s ChatGPT Astra recently surpassing the latest Claude model across numerous benchmarks, unilateral restraint carries an immediate competitive cost. Whether any peer accepts embedded evaluators, or whether Anthropic’s biggest backers keep signing checks while it slows, is the question the next earnings cycle will answer.
Contact [email protected] for any questions or corrections.








