FT: Anthropic Used AI Risk Terms 8x More Than OpenAI
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic used AI risk-related terms 8x more frequently than OpenAI in its 2026 official statements and posts, per an FT analysis by Clara Murray.
- Financial Times framed the disparity as evidence Anthropic 'may have talked itself into an export ban,' positioning the company's safety-first messaging as a regulatory liability.
- OpenAI, by contrast, used such risk language far less often, suggesting a deliberate stylistic divergence in how the two labs publicly frame advanced AI's dangers.
- A companion TechCrunch piece by Lorenzo Franceschi-Bicchierai argued that export controls historically fail, tracing a lineage from PGP encryption through spyware controls to a system it calls 'Mythos'.
- The two pieces together raise an unresolved question for U.S. AI policy: the same rhetoric that signals responsibility to safety-minded audiences is the rhetoric most likely to trigger restrictions.
Why it matters: Anthropic's safety-first posture was once a reputational asset; the FT data suggests it is now a direct regulatory liability, with potential export-control consequences for a company already valued at $950 billion. The TechCrunch historical analysis undercuts the likely remedy: if PGP and spyware export rules are the template, restrictions triggered by Anthropic's rhetoric may achieve little beyond symbolic punishment of the most safety-vocal lab.
