The company that spent a year screaming about AI as an existential threat has discovered that the US government was listening — and now it's paying the price.
An analysis by the Financial Times puts hard numbers on Anthropic's risk-obsessed messaging. Across its public statements in the past year, the company averaged five words per thousand talking about risk, regulation, or restrictions — including 336 mentions of "risk" and 121 mentions of "safeguard." That's nearly eight times the rate of OpenAI and Sam Altman, who clock in at just 0.6 per thousand words.
The disparity has earned Anthropic's safety PR a reputation as self-serving hype. Sam Altman called the strategy out in April on a podcast: "This is obviously an incredible marketing tactic — running around shouting, 'We built a bomb, and now we're going to drop it on your heads!'"
But Anthropic's "Oppenheimer-style" marketing appears to have backfired spectacularly. After major shareholder Amazon — motivated by its own capital interests — tipped off the White House about a security vulnerability in Fable 5, the administration issued a stunning 90-minute physical disconnection ultimatum. The irony? Because Anthropic had spent months telling politicians that AI risks were comparable to nuclear weapons and that Mythos possessed "cyber-weapon-grade" destructive power, the White House found the punishment entirely reasonable. Dario Amodei's carefully cultivated fear machine handed his critics the perfect excuse to pull the plug.