Anthropic, the AI research firm behind Claude, pulled the plug on its Fable 5 and Mythos 5 models late Friday night, just days after making the systems available to customers. In a brief post to its user community, the company explained that a U.S. Commerce Department order required the immediate suspension of any export of the two models, effectively banning their use outside the United States.
The directive, described by Anthropic as an "export control" measure, stems from concerns that the models could be exploited through a narrow jailbreak. According to an Axios report, an administration official said the jailbreak enables users to bypass classifier‑based safeguards designed to block prompts about cybersecurity, chemistry, and biology. The official suggested the government needed a pause to "harden" the national‑security apparatus against such threats, a process that could be completed in the next few weeks.
Government directive and national‑security concerns
Anthropic’s Friday announcement stressed that the Commerce Department provided only "verbal evidence of a potential narrow, non‑universal jailbreak" that allowed Fable 5 to review a specific codebase for software flaws. The company said it has seen the jailbreak used to find "minor" and "relatively simple" vulnerabilities, noting that similar capabilities exist in other publicly available models, such as GPT‑5.5.
"The only way for us to ensure compliance with that government order in the immediate term is that we must abruptly disable Fable 5 and Mythos 5 for all our customers," the post read. Anthropic added that the decision does not affect any of its other offerings, which remain accessible to users worldwide.
While the exact nature of the jailbreak remains undisclosed, the administration’s concern appears to center on the possibility that adversaries could leverage the models to accelerate the discovery of exploitable software bugs or to generate instructions for creating harmful biological agents. By imposing export controls, the government aims to limit the models' exposure to foreign actors while it works on strengthening defensive measures.
Industry observers note that the rapid shutdown underscores the growing tension between AI innovation and regulatory oversight. Companies that push the frontier of language‑model capability now face a tighter compliance landscape, where even a single reported vulnerability can trigger swift government action.
Anthropic’s response was swift. Within hours of receiving the directive, the firm disabled the two models across its entire customer base. The company’s engineering team reportedly coordinated with the Commerce Department to confirm that no further data or model weights had left U.S. borders.
Analysts predict that the pause could delay Anthropic’s roadmap for advanced AI products, especially as competitors race to deploy increasingly powerful systems. However, the firm’s leadership emphasized its commitment to safety and regulatory alignment, suggesting that future releases will incorporate hardened safeguards before any broader rollout.
For now, customers who signed up for early access to Fable 5 and Mythos 5 find their accounts locked, with no timeline provided for reinstatement. Anthropic has pledged to keep stakeholders informed as the situation evolves and as the government finalizes its hardening plan.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.