The U.S. government briefly pulled Anthropic’s Fable 5 from public access in mid-June after researchers found ways to bypass its safety checks. Nearly three weeks later, Anthropic reopened Fable 5 after the Commerce Department lifted export controls; the model now runs with tighter filters and limits on use while the company and regulators review risks.
Anthropic added a stronger safety filter that blocks the reported cybersecurity exploit more than 99% of the time and gives users a safe fallback answer. The company warned the filter might sometimes block harmless coding requests, though most programming work should still work normally. Paid users face reduced weekly caps for now as part of a cautious phased return.
A new element is formal pre-release access the U.S. gets to future Anthropic models, signaling closer coordination between AI labs and government reviewers before broad public rollouts. That suggests Washington wants a seat at the table for frontier releases, shifting how major models are deployed internationally.
Is this a start of real regulation or a one-off pause? It looks like early-stage control: governments are asserting influence, but the rules, gates, and processes remain unsettled and likely to evolve as new models emerge. Washington wants a seat at the table for frontier releases, shifting how major models are deployed internationally.
THE NEW NORM MAY BE PRE-LAUNCH GOVERNMENT GATES — AND AI COMPANIES WILL HAVE TO NAVIGATE THEM.
Sanjay Sahay
Have a nice evening.