The latest
The US government has lifted its restrictions on foreign access to Anthropic’s Fable 5 and Mythos 5 models after the company resolved the administration’s safety concerns — but the concessions that secured the reversal are largely a mystery, Bloomberg columnist Parmy Olson wrote, arguing the opacity leaves the market’s deepest fear intact: that American AI models can be ripped away without notice.
Details
• The ban was triggered when Amazon told the administration that certain safeguards on Fable could be circumvented, according to Bloomberg.
• Anthropic blocked the workaround using classifiers — smaller AI systems that detect requests involving prohibited cybersecurity tasks — with backing from the Commerce Department’s Center for AI Standards and Innovation (CAISI). It has disclosed little beyond the broad mechanics of the fix.
• Commerce Secretary Howard Lutnick said the company agreed to collaborate on “protocols and standards” for upcoming models, alongside tech firms including Amazon and Microsoft.
• A person close to the company said Anthropic will station staff at CAISI’s offices and give the agency computing power and its own safety testers for deeper analysis.
• More than 100 cybersecurity experts signed an open letter calling the export ban excessive, and tech executives and investors warned it would hand Chinese AI developers time to grab market share.
• OpenAI CEO Sam Altman said he disliked regulators “picking the customers” after releasing GPT-5.6 to a short list of government-approved partners, Reuters reported.
• Olson wrote that the ban deeply unnerved European companies, fueling calls to invest in homegrown AI rather than depend on American models.
What to watch
Olson argues the coming security framework should be published while it is still being drafted — not after it is settled — so the public can tell whether models are vetted on legitimate national security grounds or political favoritism. Until the triggers for any future blockade are spelled out, she warns, the uncertainty hanging over the AI market will only deepen. The next test is whether the standards Anthropic, its rivals and CAISI are drafting ever see daylight.
