Anthropic Shuts Off AI for 200M+ Users Amid US Jailbreak Risk Order

For cybersecurity professionals, the abrupt disabling of Fable 5 and Mythos 5 following a US export control order marks a critical inflection point: frontier AI models are now being evaluated as potential cyber weapons. The government's concern—a jailbreak that could automate vulnerability discovery—raises urgent questions about model security, threat modeling, and the ethics of vulnerability disclosure. This incident shows that AI’s role in offensive cyber operations is no longer theoretical.

On Friday evening, June 13, 2026, Anthropic abruptly announced the disabling of its most advanced AI models—Fable 5 and Mythos 5—for all users, a move prompted by a US government export control order barring foreign national access. The action, which affects an estimated hundreds of millions of users globally, marks a dramatic escalation in the Trump administration's clampdown on AI technologies, shifting from hardware chips to software restrictions. The order, delivered at approximately 5:21 p.m. ET, cites undisclosed national security concerns but is understood by Anthropic to center on a narrow jailbreak that could allow Fable 5 to identify software vulnerabilities. Anthropic’s decision to globally disable rather than attempt selective compliance underscores the difficulty of enforcing nationality-based access controls for cloud-delivered AI services.

“

On Friday evening, June 13, 2026, Anthropic abruptly announced the disabling of its most advanced AI models—Fable 5 and Mythos 5—for all users, a move prompted by a US government export control order barring foreign national access.

The order arrives amid a fraught history between Anthropic and the Trump administration. Earlier in 2026, the Pentagon designated the IPO-bound company as a supply chain risk after Anthropic refused to allow its models to be used for domestic surveillance and fully autonomous weapons. Anthropic sued, and two lawsuits remain pending. That blacklisting, set to take effect later this year, had already strained the relationship. However, recent signs of easing tensions across parts of the US government were abruptly reversed by this latest directive, which Anthropic says provided only “verbal evidence of a potential narrow, non-universal jailbreak” without detailed specifics.

The core national security concern revolves around a potential method to bypass Fable 5’s safeguards—a jailbreak that could enable the discovery of software vulnerabilities. While Anthropic acknowledges the government’s concern, it disputes the severity, arguing that the technique is narrow, involves known vulnerabilities identifiable by other publicly available models, and does not justify a blanket recall. “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company stated. This disagreement highlights a fundamental rift: regulators view any exploitable ML-based vulnerability discovery as a critical cyber threat, while developers like Anthropic call for proportional, evidence-based risk assessment.

The broader context is a US export control regime that has long focused on physical semiconductors and chipmaking tools to curb China’s AI ambitions. This order represents a watershed moment—the first major restriction on foreign access to a specific AI model itself. It signals that the US government now considers frontier AI models as dual-use technologies on par with weapons systems, subject to International Traffic in Arms Regulations (ITAR) or similar controls. For the cybersecurity community, the implications are twofold. First, it validates fears that advanced AI can accelerate vulnerability research, lowering the barrier for malicious actors to find and exploit zero-days. Second, it raises profound questions about model security testing and responsible disclosure, as even internal red-teaming could be recast as a national security risk.