Anthropic Reaches Deal With Trump Administration to Restore Access to Fable AI Model
The Wall Street Journal ● Covered by 3 sources
Anthropic struck a deal with the Trump administration to bring back access to its Fable AI model. Amazon researchers had found ways around its safety guardrails, so Anthropic's fixing that as part of the deal.
Based on reporting by The Wall Street Journal — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic and the Trump administration have hashed out an agreement to restore access to Fable, the company's AI model that got pulled after researchers at Amazon figured out how to slip past its safety guardrails. Access started coming back online today, according to details of the arrangement.
The workarounds themselves haven't been fully detailed publicly, but they were apparently significant enough to trigger a shutdown and a negotiation involving federal oversight. That's not a small thing. Model access getting cut off, then restored through a government-brokered deal, signals just how seriously safety bypasses are now being treated inside Washington, especially for a lab like Anthropic that has built its reputation on being the safety-conscious alternative to more move-fast competitors.
The Center for AI Standards and Innovation is expected to play a role in this agreement. CAISI is one of the newer government units built specifically to test and evaluate AI systems, and its involvement here suggests the fix isn't just Anthropic patching a hole and calling it a day. There's a verification layer now, a government body checking that whatever Anthropic did to close the gap actually holds up.
What stands out is the fact that it was Amazon, a company with its own deep AI ambitions and a major investor in Anthropic, whose researchers found the exploit in the first place. Whether that discovery came through adversarial red-teaming, internal testing, or something closer to an accident isn't clear from what's been shared. But it does show that even well-resourced, safety-focused labs are still getting caught out by people probing their own products, sometimes from inside the same corporate family.
Anthropic hasn't laid out exactly what changed under the hood, only that the workarounds are being addressed and that restoration is underway. For a company that has leaned hard into the narrative of being the responsible AI lab, getting a model pulled and then having to negotiate its return through a federal standards agency is the kind of episode that tests whether that reputation is earned or just marketing.
My take — AI-written commentary, not fact-checked reporting
This is exactly the kind of incident that should worry people more than it seems to, because Amazon, an investor with skin in Anthropic's success, found the hole before any adversarial outsider did. Government-brokered restorations are becoming the new normal for frontier labs, and I'd rather see CAISI's fingerprints on these deals early than find out later that a bypass slipped through unnoticed. Safety-first branding means nothing if the safeguards can be talked around by researchers sitting one floor away from the money.”
Read more about this at: The Wall Street Journal