The United States government issued an export control directive ordering Anthropic to suspend access to its two most capable Artificial Intelligence models, Claude Fable 5 and Claude Mythos 5, for all foreign nationals worldwide, citing national security concerns and forcing the company to disable both models entirely for all customers as the only feasible path to compliance.
Anthropic received the directive at 5:21 PM Eastern Time on June 13, just three days after the models launched, with the order coming from the Commerce Department under the authority of Commerce Secretary Howard Lutnick. Because the restriction covered foreign nationals both inside and outside the United States, including non-citizen employees working at Anthropic itself, the company said selective compliance was technically impossible and had no choice but to pull both models globally. Access to all other Anthropic models, including Claude Opus 4.8, was not affected. Anthropic said it disagreed with the directive but was complying while working to restore access as soon as possible.
The models at the centre of the dispute sit at the top of Anthropic’s capability hierarchy. Mythos 5, which has not been made publicly available, was initially restricted to approximately 50 vetted organisations including Amazon, Apple, Google, Microsoft, and CrowdStrike through a controlled programme called Project Glasswing, specifically for defensive cybersecurity work, after the model demonstrated the ability to identify flaws in every major operating system and web browser it tested, including vulnerabilities that had gone undetected for decades. Fable 5, released three days before the directive, was designed as a commercially accessible version of Mythos with guardrails blocking its most sensitive cybersecurity and biotechnology capabilities, and immediately became the most capable Artificial Intelligence model available to the public according to independent benchmark testing. The US government informed Anthropic it had become aware of a method of bypassing Fable 5’s safeguards. Anthropic reviewed what it described as a narrow, non-universal jailbreak that it said demonstrated only a limited ability to identify minor, previously known vulnerabilities already discoverable by other publicly available models including OpenAI’s GPT-5.5, and argued the same jailbreak technique could not be used as grounds to recall a model deployed to hundreds of millions of people without effectively halting all new frontier model deployments across the Artificial Intelligence industry.
The episode has exposed the tension at the centre of Anthropic’s commercial and policy positioning. When Anthropic first previewed Mythos, it described the model’s cybersecurity capabilities as uniquely dangerous and restricted access accordingly. Critics, including OpenAI Chief Executive Sam Altman, argued this framing amounted to marketing that overstated risk. When the government ultimately took those warnings at face value and issued a recall, Anthropic found itself arguing the opposite, that the model was safe enough for broad commercial deployment and the jailbreak did not meet the bar for the action taken. David Sacks, an adviser to President Trump, said on social media that the government had notified Anthropic of the jailbreak before issuing the directive and that the company took no action to address it. The development arrives at a commercially sensitive moment for Anthropic, which had confidentially filed for a public listing shortly before the incident, with the shutdown of its two most capable models creating uncertainty heading into what the company had expected to be a period of strong commercial momentum.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
