The US government abruptly restricted global access to Anthropic's Claude Fable 5 for three weeks. A security researcher had found a prompt that could bypass Fable 5's safety guardrails and identify software vulnerabilities. This incident exposed a critical lack of industry standards for AI jailbreak severity, which Anthropic is now addressing.
How We Got Here
Anthropic launched Fable 5 on June 9, but the US government imposed export controls just three days later, on June 12. This rapid action followed an Amazon research report demonstrating Fable 5's vulnerability to a prompt that bypassed safety measures.
The Numbers
- Global access for Fable 5 resumes July 1 across Claude Platform, Claude.ai, Claude Code, and Claude Cowork.
- Until July 7, Pro, Max, Team, and Enterprise users will receive Fable 5 access under 50% of their existing usage limits.
- Anthropic developed a new safety classifier that now blocks the reported bypass technique in over 99% of cases.
- The US Department of Commerce's Center for AI Standards and Innovation (CAISI) tested and approved Anthropic's enhanced safeguards.
What Happens Next
🇮🇳 Why This Matters for India
For Bangalore's AI product managers and engineers, restoring Fable 5 access means unblocked development, but the new classifier's higher false positives could slow specific coding workflows.
The Take
The real takeaway here isn't just the fix, but how quickly national security concerns can halt commercial AI deployment. Anthropic's new false-positive trade-off shows the industry now prioritizes a "safer" but clunkier tool over an unfettered one, setting a precedent.
Source:
MediaNama ↗