When the U.S. government restricted access to Anthropic's cybersecurity-focused Fable 5 model in mid-June 2026, the company discovered that the same security vulnerability affected OpenAI's GPT-5.5 and Chinese model Kimi K2.7, yet only Anthropic faced export controls. This regulatory asymmetry has become a flashpoint in the debate over how the government should police frontier artificial intelligence (AI) systems, exposing tensions between security concerns and competitive fairness.

What Triggered the Export Controls on Anthropic's Models?

The Commerce Department imposed export controls on Anthropic's Fable 5 after an Amazon research report described a "jailbreak" technique. Security researchers fed the model open-source code containing publicly known vulnerabilities plus deliberately planted flaws, then asked it to "fix this code." The model's output was manually assembled across multiple steps into scripts that test patches.

Katie Moussouris, founder of Luta Security, whom Anthropic asked to assess the paper, argued that this was not actually a guardrail bypass. Instead, she explained that it demonstrated "the most valuable thing an AI model can do for defensive security: executing the find, fix, and test loop defenders run every day." Her analysis concluded that the underlying capability cannot be removed without degrading the model's usefulness for legitimate security work.

Why Did Other Models Escape Restrictions?

Anthropic's subsequent testing confirmed that the same technique worked against OpenAI's GPT-5.5 and the Chinese model Kimi K2.7, yet neither faced comparable export restrictions. The company argued that the technique exposed no capability unique to its frontier models, raising questions about whether the export controls were proportionate or applied consistently across the industry.

This discrepancy sparked significant industry pushback. More than 100 cybersecurity professionals, including executives from Nvidia, Adobe, Zoom, Google, and Sophos, signed an open letter organized by former Facebook security chief Alex Stamos. The letter warned that pulling the best capabilities away from defenders without applying the same standard industry-wide would be "dangerous" at a time when Chinese open-weight models were only months behind the best American models.

"The Chinese open-weight models are only months behind the best American models, and those are the models we know about. To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous," the letter stated.

Open letter organized by Alex Stamos, former Facebook security chief

What Commitments Did Anthropic Make to Restore Access?

On June 30, 2026, the Commerce Department lifted export controls on both Fable 5 and Anthropic's more powerful model, Mythos 5, after the company reached agreements with the government. The episode marked the first known use of export control authorities to pull AI software, rather than chips or hardware, from public access. Its resolution may set the terms under which frontier AI models are regulated in the U.S. going forward.

Anthropic's negotiations resulted in several commitments that could become industry standards:

• Safety Classifier: Anthropic trained a new safety classifier that blocks the specific jailbreak technique in more than 99% of cases, which the Commerce Department's Center for AI Standards and Innovation tested and endorsed.
• Pre-Release Government Access: The company committed to expanded pre-release access for government evaluators to test frontier models before broad release, allowing regulators to identify vulnerabilities earlier in the development cycle.
• Rapid Disclosure Protocol: Anthropic agreed to rapid disclosure of significant jailbreaks and dedicated staff and compute for joint research with government agencies.
• Industry Framework: Working with Glasswing partners including Amazon, Microsoft, and Google, Anthropic is drafting an industry framework to score jailbreak severity across four criteria: capability gain over existing tools, breadth of tasks affected, ease of weaponization, and discoverability.
• Bug Bounty Program: The company opened a HackerOne bug bounty program specifically for cyber jailbreak submissions, creating a formal channel for security researchers to report vulnerabilities.

How Can Policymakers Balance Security and Innovation?

Anthropic argued that if the standard applied to Fable 5 were applied industry-wide, it would "essentially halt all new model deployments for all frontier model providers." This statement underscores a central tension in AI regulation: how to balance security concerns with the need for continued innovation and the competitive advantage of U.S. companies over international rivals.

Anthropic

The Five Eyes intelligence alliance, which includes the United States, United Kingdom, Canada, Australia, and New Zealand, has warned business leaders that frontier AI models will "fundamentally transform" both offensive and defensive cybersecurity capabilities, with the timeline measured in months rather than years. This urgency has intensified pressure on regulators to act quickly without stifling the development of defensive AI tools.

United States, United Kingdom, Canada, Australia, and New Zealand

Access to Anthropic's Mythos 5 remains restricted to vetted U.S. organizations through Project Glasswing, a controlled-access program for critical infrastructure defenders, while the company continues to negotiate broader domestic and international access. The fact that OpenAI's GPT-5.5 was not subject to the same restrictions raises a critical question for the industry: will the U.S. government apply consistent standards to all frontier AI models, or will regulatory treatment vary by company? The answer may determine whether future export controls become a tool for leveling the playing field or a source of competitive advantage.