Parmy Olson
Parmy Olson argues in Bloomberg Opinion that the lifting of U.S. restrictions on foreign access to Anthropic’s “Fable 5” and “Mythos 5” models has not removed the deeper concern among AI customers outside the United States: What happens if Washington can suddenly cut off access to American AI models?
According to Olson, closed-door negotiations between Anthropic and the U.S. government succeeded in reversing the restrictions after the company addressed the Trump administration’s safety concerns. But the troubling part, she argues, is that the terms of the agreement remain largely unclear.
Olson writes that the ban began after Amazon warned the administration about concerns that some safeguards in Anthropic’s “Fable” model could be bypassed. Anthropic then built a fix using smaller classifier systems designed to detect requests involving prohibited cybersecurity tasks, with support from the Commerce Department’s Center for AI Standards and Innovation.
But Olson’s central point is that a technical fix is not enough. She argues that rules for testing AI models and imposing export restrictions should be public and transparent, not handled through private, case-by-case negotiations. Without that clarity, customers cannot know whether AI systems are being restricted for legitimate national security reasons or because of political pressure and favoritism.
The concern, she notes, is not limited to Anthropic. OpenAI customers may face the same fear if access to U.S. models can be limited by government decision. Olson cites OpenAI CEO Sam Altman’s objection to the idea of regulators “picking the customers.”
Olson concludes that AI governance should not be improvised. Clear public standards are necessary to preserve trust in the market. Otherwise, European and other foreign companies may accelerate efforts to build or adopt non-American alternatives, fearing that U.S. AI tools could be pulled away without warning.
Bloomberg Opinion.