Dario Amodei answered investor Gavin Baker in public after Baker argued that the Anthropic chief executive’s warnings had fed American hostility toward AI. Amodei put the cause somewhere else and described a trust crisis that runs far wider than the industry. Along the way he conceded that AI companies, his own included, have not delivered what they promised.
Key Takeaways
- Amodei frames the current backlash as a crisis of trust rather than a messaging problem.
- He rejects the claim that his public record is lopsided and points to one major essay on risks and one on benefits.
- He accepts the harshest criticism aimed at the sector: the promises made were never delivered.
The Charge Gavin Baker Levelled at Anthropic
The exchange started with Baker, who argued on the All-In podcast and then on X that Amodei’s repeated warnings about AI dangers had helped turn American opinion against the technology, and against data centers in particular.
That accusation lands harder on Anthropic than it would on any other lab. The company spent two years positioning itself as the frontier lab that discusses risk before it discusses capability, and that stance now works as a commercial asset as much as an ethical line. Being named as the reason infrastructure gets blocked attacks the position at its root.
The reply came on August 15 as a two-part thread posted on his own account. Amodei opens by thanking Baker for the exchange, notes that he rarely spends time on social media, then takes the two criticisms apart one after the other. The first concerns the balance of his public messaging, the second concerns regulation.
On balance he pushed back without hedging. His messaging has been about equally split between risks and benefits, with one major essay devoted to each side, and the interviews where he discusses danger carry the upside too. The souring public mood, on his reading, did not start there.
The timing matters because American anxiety about work already shapes a large slice of the rejection. The survey Anthropic commissioned on job fear found a majority already worried, well before AI turned into a campaign theme.
The Undelivered Promise Amodei Owns
The most unusual passage in the thread is the one where Amodei accepts a criticism instead of answering it. He calls the fairest charge against the sector the fact that AI companies, Anthropic among them, have not delivered the transformation they announced. A frontier lab chief executive writing that down remains rare.
His phrasing is blunt. Actually curing cancer counts for more than promising that AI will cure cancer. The distinction sounds obvious on paper, yet it maps neatly onto the gap between how the industry has talked since 2023 and what a working professional sees on screen.
The diagnosis he draws from it reaches well past AI. Ordinary people do not trust companies, governments or the tech industry, and they assume by default that someone is cooking up a fresh way to cheat them. The backlash inherits that accumulated suspicion rather than manufacturing it.
For teams shipping AI inside a company, the consequence is practical. The burden of proof moves from narrative to measurable result, and a rollout that cannot show its gain becomes much harder to defend in a budget meeting. The same pressure now applies to infrastructure, where the data center Anthropic rented from Riot Platforms gets negotiated in a local climate where every megawatt needs a justification.
There is a less generous reading of the admission. Conceding that the sector underdelivered also defuses the alarmism charge, since a company that failed to produce the promised upside can hardly be accused of having oversold the downside either. The thread leaves that tension open.
More articles on Horizon
- Anthropic Left Its Bioweapon Filter Off for 11 Months
- OpenAI Shuts Down Its Catastrophic Risk Team
- Qwen3.8-27B Is Now Free to Download and Modify
What the Regulatory Line Means for Rival Labs
The second half of the reply covers regulation, and that is where the competitive effect sits. Amodei refuses the binary framing and spells out the line he defends: slow down frontier AI companies while giving smaller competitors an advantage. The Californian transparency bill Anthropic supports follows that logic.
The move is clever because it inverts the standard accusation. Large labs are routinely charged with pushing rules that lock the market behind them, a capture argument aimed at OpenAI as much as at Anthropic. Claiming a rule that penalises the biggest players first, yourself included, is one way out of that trap.
Transparency obligations do cost a lab training frontier models far more than they cost a vendor building on an API. If the Californian template spreads, the constraint concentrates on five or six companies while the application layer inherits a more readable environment without paying for it.
Anthropic is already several steps into that ground. The agreement with Gavin Newsom to supply Claude to state services at half price installs the lab as an institutional counterpart exactly as the framework takes shape, which buys more than a public statement ever could.
Rivals now have to place themselves. As long as the debate pitted acceleration against caution, each lab could pick a side at no immediate cost. A chief executive who publicly grants that his sector failed to deliver changes the terms of the conversation and makes everyone else’s silence louder than it was last week.
Follow the story on Horizon.
