How a Code Audit Jailbroke Anthropic and Triggered the First Global AI Shutdown

The 90-Minute Compliance Deadline

At 5:21 PM Eastern on June 12, 2026, a letter from US Commerce Secretary Howard Lutnick landed at Anthropic's headquarters . The directive was sweeping: Anthropic was prohibited from exporting its Fable 5 and Mythos 5 models to any destination worldwide, or from granting access to any foreign national regardless of location — including foreign nationals employed by Anthropic itself . Lutnick threatened criminal and civil penalties for noncompliance .

Because the directive covered "any foreign national anywhere," Anthropic concluded it had no way to verify every user's citizenship in real time. The only path to compliance was to disable both models for all customers worldwide. "The net effect of this order is that we must abruptly disable access to these models for all users," the company stated in its official blog post . Anthropic had roughly 90 minutes to comply before the letter took full legal effect .

Anthropic's "Misunderstanding" Defense

Anthropic immediately characterized the government's action as based on a "misunderstanding" . The company argued that the jailbreak capability — asking the model to audit code for vulnerabilities — was not unique to Fable 5. It already existed in OpenAI's GPT-5.5, in Chinese models like Kimi 2.7, and in virtually every other deployed frontier model . Anthropic also noted that no testers had yet found a "universal jailbreak" — a method that could broadly bypass the model's safeguards — and that the disclosed findings revealed only minor, previously known vulnerabilities that provided no Mythos-specific uplift .

The Backstory: A Frayed Relationship with the Administration

Anthropic's relationship with the Trump administration had been deteriorating long before the June 12 directive. The company had publicly resisted providing its AI models for Pentagon and military intelligence applications, including refusing a Department of Defense contract for Claude . This put Anthropic squarely at odds with the administration's push for AI dominance and military AI integration. Defense Secretary Pete Hegseth later "claimed vindication" after the export order, arguing it proved the administration's concerns about Anthropic's model security were justified .

Wall Street Reacts: The Hong Kong Bans

The export control order sent immediate shockwaves through the financial industry, particularly around the sensitive question of deploying US AI models in Hong Kong — a jurisdiction caught between US export controls and China's regulatory reach.

Goldman Sachs was the first to act. In late April 2026 — well before the June 12 order — Goldman instructed its Hong Kong bankers to stop using Anthropic's Claude models after a weeks-long unexplained access blackout. The decision came after Goldman consulted Anthropic and took a strict interpretation of its service contract, concluding that Hong Kong-based employees should not have access to any Anthropic products . Other AI tools, including ChatGPT and Google Gemini, remained available to Goldman's Hong Kong staff .

JPMorgan Chase followed on June 18, 2026 — six days after the Lutnick directive. JPMorgan barred its Hong Kong staff from selecting Claude models from the bank's internal list of approved large language models . The Financial Times reported the move reflected heightened scrutiny of AI use outside the US . JPMorgan's decision was driven by concerns over the language in Anthropic's licensing agreement's terms amid escalating US-China technology tensions .

Notably, neither ban applied to US-based employees.

The Broader Fallout

The consequences of the June 12 directive extended far beyond Wall Street and Hong Kong:

• Microsoft blocks Claude access: Microsoft moved to restrict access to Anthropic's models across its platforms in response to the regulatory turmoil .
• UK and EU pushback: European regulators and the UK government expressed concerns about the extraterritorial reach of US export controls on AI, questioning the precedent of a single government unilaterally shutting down a globally deployed commercial AI model .
• Chinese AI stocks surge: Chinese AI companies saw stock rallies as the US action against Anthropic was interpreted as a competitive blow to US AI leadership, boosting sentiment for domestic Chinese AI alternatives .
• Anthropic's IPO thrown into uncertainty: Anthropic's IPO plans — already under discussion before the crisis — were complicated by the export order, the loss of two flagship products, and the reputational damage .

The age of software export controls on AI has begun. The question now is whether this was a one-time overreaction, as Anthropic insists, or the first move in a new era where the world's most powerful AI models are treated as strategic assets — no different from missiles or nuclear secrets.

As of June 2026, Fable 5 and Mythos 5 remain offline globally, with no announced timeline for their return.