Enterprise customers of Anthropic and OpenAI still worry about U.S. kill-switch risk after the closed-door deal that reversed Anthropic’s export ban left its terms largely undisclosed, according to Bloomberg Opinion. The ban was triggered when Amazon flagged that a safeguard on Anthropic’s Fable model could be circumvented, and Anthropic fixed it using classifiers, smaller AI systems that detect prohibited cybersecurity requests, with backing from the Commerce Department’s CAISI. Anthropic also agreed to collaborate with the government and rival firms including Amazon and Microsoft on security protocols for future models, though those standards have not been made public. The 18-day ban unnerved European companies, prompting calls to build homegrown AI they can’t have ripped out from under them without warning.
