U.S. Lifted Anthropic Curbs After Fable 5 Fell to 3 of 12 Tasks

3 articles · Updated · UPI News · Jul 21

June 30 brought a reversal of the Commerce Department's June 12 order after Anthropic tightened guardrails on Claude Fable 5 and Mythos 5 so sharply that benchmark performance collapsed.
One test found the revised Fable 5 completed just 3 of 12 previously routine tasks, showing the restrictions were lifted only after the models were materially weakened.
The order had barred all foreign nationals — including Anthropic's own noncitizen staff — from using the models, and Anthropic responded by shutting access for everyone because it could not quickly verify every user's nationality.
The episode marked the first known use of U.S. export controls to block foreign access to an AI service, raising questions over whether model outputs count as exports and whether a global, open-ended order exceeded Commerce's legal authority.
Anthropic did not challenge the directive in court, instead negotiating with Washington — a stance aligned with CEO Dario Amodei's call for government power to block deployment of dangerous frontier AI models.

Sources

Left50%

Center50%

UPI News1d ago

U.S. Commerce Dept. Hobbles Anthropic AI Models, Raises Export Control Questions

CNBC1d ago

The Fed rang the alarm about Anthropic's Mythos AI — but went without it

The Conversation1d ago

Legally or not, the US government is controlling global access to the world's most powerful AI

8 Sources

With API access now an 'export,' could the U.S. government shut down any AI service overnight?

After a government order crippled its AI, can Anthropic balance ultimate power with absolute safety?

Did Anthropic’s CEO invite government control to ensure safety, or to outmaneuver his rivals?

The June 2026 Anthropic Fable 5 Export Controls: Causes, Industry Disruption, and the Future of AI Governance

Overview

In June 2026, the U.S. government imposed strict export controls on Anthropic’s advanced AI models after Amazon researchers discovered a major jailbreak vulnerability in Claude Fable 5, which could bypass security and reveal sensitive information. Anthropic responded by suspending global access to its models, causing confusion and operational disruptions for enterprises. Working closely with the Department of Commerce, Anthropic quickly developed new safeguards and used Amazon Bedrock’s compliance tools to restore access. The incident highlighted the risks of relying on single AI vendors and led to industry-wide efforts to improve AI safety and create shared standards for handling vulnerabilities.

...