US Government Restricts Access to Anthropic's Fable and Mythos AI Models

The U.S. government has ordered AI company Anthropic to restrict exports and block foreign access to its new advanced models, Fable 5 and Mythos 5, citing national security concerns. The directive, which reportedly stemmed from concerns about the models' automated hacking capabilities and potential for jailbreaking, has forced Anthropic to take the models offline globally and has sparked a debate over AI regulation and US competitiveness.
US Government Restricts Access to Anthropic's Fable and Mythos AI Models

US Government Restricts Access to Anthropic’s Fable and Mythos AI Models The clash between Anthropic and the U.S. government over its newest AI models began as a technical dispute and rapidly escalated into a global test of political power and AI governance.

How the crisis unfolded

Anthropic launched its cybersecurity-focused Claude Mythos 5 to a restricted group in early June, warning it was “too dangerous” for broad release because of its ability to find high‑severity vulnerabilities. Days later, it released Claude Fable 5, a Mythos‑grade model with extra safeguards, to the general public on June 9.

On June 12, after Amazon researchers demonstrated a “fix this code” prompt that could coax Fable 5 into surfacing software flaws, Amazon CEO Andy Jassy escalated the issue directly to senior Trump administration officials. That evening, the Commerce Department sent Anthropic an export‑control directive barring “any foreign national” from accessing Fable 5 and Mythos 5, including Anthropic’s own staff. With about 90 minutes’ notice, Anthropic took both models offline worldwide “to ensure compliance,” arguing the discovered jailbreak was narrow and already patched.

Competing narratives

The White House framed the move as a national‑security measure, reportedly driven by fears that Chinese‑linked actors could bypass Fable’s guardrails or access Mythos via foreign partners. The order marked “the first time US export controls have been used to control access to an AI model in this way.”

Anthropic and many security experts countered that the exploit was routine and not unique to Fable. An open letter signed by about 100 cybersecurity professionals warned that “this action has taken the best models away from defenders… without any real risk to justify it.” Critics described the directive as hasty and opaque, suggesting it reflected “personality differences” and mistrust between Anthropic and the administration as much as technical risk.

Fallout at home and abroad

The shutdown immediately disrupted users: one UK founder saw Fable vanish “mid‑task” during a critical code review, reinforcing his view that businesses must not rely on a single U.S. AI provider. More broadly, the episode showed that Washington “can decide who may use the world’s most important technology,” underscoring America’s role as gatekeeper to frontier models and compute.

Abroad, the ban fueled calls for “AI sovereignty.” European startup Mistral, which champions open‑weight models that customers can run on their own infrastructure, seized on the incident as proof that foreign firms are vulnerable to U.S. political whims. Commentators warned that cutting access to Mythos and Fable was “a gift” to Chinese competitors and other non‑U.S. labs that can pitch themselves as more reliable partners.

A search for new rules

Over the following days, Anthropic’s senior technical staff flew to Washington for crisis talks with Commerce officials, attempting to prove their safeguards and negotiate a path to restoration. The dispute has spurred U.S.–EU discussions of a “trusted partner” scheme for controlled access to cutting‑edge models, and intensified industry debates over whether brittle API guardrails are viable, or whether deeper controls and new testing benchmarks for jailbreaks are needed.

Across perspectives, one point of agreement is emerging: powerful “dangerous” AI models are coming regardless, and ad‑hoc, opaque crackdowns are unlikely to contain them for long.

Continue reading https://foxvector.com/stories/019eebd7-45b7-0487-73be-1531d45af809

Write a comment