The US government just shut down Anthropic's best model for 3 weeks. Your vendor risk model just broke.

The US Department of Commerce imposed export controls on Claude Fable 5 on June 12.

Amazon researchers found a jailbreak that let the model identify software vulnerabilities and produce exploit code.

Anthropic couldn't verify user nationality in real-time. So they suspended access for everyone. Globally.

For three weeks, every enterprise running Fable 5 had zero access. No workaround. No fallback plan.

Here's what your CISO missed:

The jailbreak wasn't even that dangerous. Anthropic's own testing showed that Opus 4.8, GPT-5.5, and Kimi K2.7 could all identify the same vulnerabilities. The government acted anyway.

Because the capability exists. Not because it was unique.

This is the new regulatory reality. The US government will shut down an AI model based on what it *can* do, not what it *did* do.

Anthropic has now built a new safety classifier that blocks the bypass in 99%+ of cases. They're also working with Amazon, Microsoft, and Google on a shared framework for scoring jailbreak severity.

But here's the part that should keep you up at night:

Anthropic committed to pre-release government access for future models. That means the US government gets to test your AI vendor's models before you do.

Audit your AI vendor contracts today. If your SLA doesn't cover government-imposed access suspension, you're exposed. The next model shutdown could last longer than three weeks.

SOURCE: https://www.anthropic.com/news/redeploying-fable-5
VERIFIED: Anthropic official blog, TechCrunch, Reuters
SIGNAL: This is the first time the US government used export controls to shut down a consumer AI model. Every enterprise running frontier models just learned that their vendor can be taken offline by regulatory action with zero notice.
---