Anthropic used risk or regulation words five times per 1,000 in 2026; OpenAI used them eight times less. Last week, Washington barred foreign nationals from using Anthropic's newest models, Mythos and Fable. That's not a coincidence.
The Numbers That Killed the Deal
FT analyzed official statements, social media posts, and articles from Anthropic CEO Dario Amodei. Five of every 1,000 words related to risk, regulation, or restrictions. For OpenAI and Sam Altman, that figure was 0.6 words per 1,000. The $965 billion AI company's own rhetoric painted a target on itself.
Washington's export ban targets Mythos specifically. Some technologists point to Anthropic's repeated warnings about Mythos's potential societal harm as the trigger. You don't tell regulators your product is dangerous and then expect open global distribution.
Cause or Convenient Scapegoat?
I'm not saying Anthropic should have lied. But the timing stinks. The company spent 2026 amplifying doomsday scenarios. Now the government hands down an export ban that conveniently aligns with that narrative. Critics accuse Anthropic of talking itself into regulatory captivity.
OpenAI, by contrast, kept risk language minimal. Their models face no such export restrictions. The difference isn't technical capability - it's public posture. Anthropic's PR strategy may have locked the door on international markets.
What This Enables Next
Expect other AI labs to recalibrate their risk messaging. The cost of alarmism just became a $965 billion export ban. Foreign competitors won't be blocked from building similar capabilities - they'll just develop them without American oversight. Anthropic's own words may end up accelerating the very proliferation they warned about.
Source: How Anthropic may have talked itself into an AI export ban
Domain: arstechnica.com
Comments load interactively on the live page.