AI News Briefing — June 14, 2026

Anthropic abruptly disabled its Mythos 5 and Fable 5 models for all customers worldwide, just days after launch, following a US Commerce Department export-control directive. The administration reportedly cited a jailbreak that bypasses classifier-based safeguards, and Anthropic says full shutdown was the only way to comply immediately.

Anthropic said the US government believes it discovered a method to jailbreak Fable 5, prompting the order to take the model offline. The company is complying while framing the issue around a single bypass technique.

Anthropic confirmed it will disable its most advanced models after the US ordered access suspended for foreign nationals on national security grounds. The company says it received no specific details of the concern and that the safeguards could allegedly be bypassed to identify software vulnerabilities.

Anthropic cut off Fable 5 and Mythos 5 for all customers after the government ordered access blocked for all foreign nationals, including its own employees. The company says evidence of the jailbreak was provided only verbally and that the vulnerabilities were minor.

A German court ruled that Google is legally liable for false statements produced by its AI Overviews feature. The decision establishes that a company designing, training, and operating an AI system bears responsibility for damages from its generated outputs.

A detailed report recaps Anthropic disabling Fable 5 and Mythos 5 for all customers after a June 12 export-control directive that suspended access for any foreign national, including Anthropic's own foreign staff. Because real-time filtering of foreign nationals is impossible, the company shut both models down entirely while leaving other models like Opus 4.8 online.

Anthropic publicly pushed back against the shutdown order, arguing that a narrow potential jailbreak should not justify recalling a model used by hundreds of millions. The company's frustration signals tension between safety advocacy and government enforcement.

Per a Wall Street Journal report, Amazon cybersecurity research and Jassy's conversations with the White House helped trigger the export-control directive against Fable 5 and Mythos 5. Amazon's paper reportedly showed prompts that elicited information usable in cyberattacks.

Anthropic's Claude Fable 5 reportedly reached 88 percent on the hardest FrontierMath tier, a dramatic leap from Opus 4.5's sub-10 percent earlier in 2026, versus about 75 percent for GPT-5.5. The result underscores rapidly accelerating AI mathematical reasoning.