In detail
- FT reports Anthropic used risk‑related terms about 5 times per 1,000 words in 2026 vs. ~0.6 for OpenAI.
- Anthropic communications counted 'risk' 336 times, 'safeguard' 121, 'vulnerability' 128 in 2026.
- Some technologists argue Anthropic's frequent warnings helped prompt the US decision to restrict access to Mythos and Fable.
- Public debate raises questions about how company messaging affects regulation and policy responses.
Why it matters
How firms talk about AI risks can influence policymakers; public warnings may attract restrictive regulation rather than protective frameworks.
For you If your company publicly discusses AI risks, align messaging with legal and policy teams and anticipate potential regulatory blowback.