Ricardo Howard
Anthropic's Claude Fable 5 and Mythos 5 models are back online worldwide after the U.S. government lifted export controls that had shut down both systems for 18 days, the company said Tuesday.
What Changes Starting Wednesday
Global access to Fable 5 resumes across Claude.ai, Claude Code, Claude Cowork, and the broader Claude platform. Anthropic is sweetening the return: through July 7, subscribers on Pro, Max, Team, and select Enterprise tiers get Fable 5 usage covering up to half their weekly allotment before any credit deductions kick in. That's a meaningful concession given how disruptive the outage was for teams that had built workflows around the model.
Cloud reinstatement is coming but not yet scheduled. Anthropic confirmed plans to restore Fable 5 on Amazon Web Services, Google Cloud, and Microsoft Foundry, though it gave no firm date. Mythos 5, the more restricted of the two models, only regained clearance for a limited set of U.S. organizations after federal sign-off on June 26. Broader distribution, including international partners through the Glasswing program, is still being negotiated with regulators.
Why the Models Went Dark in the First Place
The June 12 directive traced back to a technique documented by Amazon researchers for coaxing Fable 5 into producing dangerous outputs tied to software vulnerability discovery. Because the export restriction applied to all foreign nationals, a category that swept in researchers on Anthropic's own payroll, the company decided the only way to comply was to pull both models offline entirely rather than attempt a narrower carve out.
Anthropic pushed back on the notion that Fable 5 was uniquely at fault. Internal testing, the company said, showed that other widely deployed models, including OpenAI's GPT-5.5, could be coaxed into generating the same flagged outputs. That detail matters for how the industry and regulators think about model-specific restrictions versus capability-wide risks that cut across vendors.
In response, Anthropic built a new safety classifier designed specifically to catch the Amazon-documented exploit. Company testing found it blocks the technique in more than 99 percent of attempts. The Commerce Department's Center for AI Standards and Innovation reviewed both the original guardrails and the new classifier and confirmed their effectiveness, according to Anthropic. Any requests to Fable 5 that still get blocked will be rerouted automatically to Claude Opus 4.8, so users shouldn't hit a dead end even during edge cases.
Washington's Involvement and the Timing Problem
Commerce Secretary Howard Lutnick said on X that his department spent roughly two weeks vetting Fable 5 with Anthropic to get alignment across the federal government and to shore up the country's competitive position in AI. White House chief of staff Susie Wiles echoed that framing on X, saying the shared goal was deploying strong technology as fast and safely as possible, a comment first reported by The Wall Street Journal.
The timing could hardly have been worse for Anthropic. The shutdown landed just as the company had filed a confidential IPO prospectus showing a $47 billion revenue run rate and a $965 billion valuation. Notably, co-founder Tom Brown, not CEO Dario Amodei, took the lead in negotiating with regulators, a shift visible in the fact that Lutnick's correspondence was addressed directly to Brown, according to CNBC.
