U.S. Commerce Secretary Howard Lutnick Orders Freeze on Access to Anthropic's Fable 5 and Mythos 5
The U.S. Department of Commerce has imposed export controls on Anthropic's latest models for national security reasons, requiring a suspension of access to Fable 5 and Mythos 5 for any foreign nationals, including foreign employees of companies.
In compliance with the directive, Anthropic has fully disabled access to both models. Fable 5, released on June 9, is a commercial version with strict filters, while Mythos 5, a 'bare model,' is limited to a few vetted red teams and defense departments under Project Glasswing.
After third parties claimed to have jailbroken the controlled Mythos 5, the Department of Commerce intervened swiftly. This move is the result of weeks of behind-the-scenes negotiations, as the government had previously attempted to block the release without success.
Source: Public Information
ABAB AI Insight
Anthropic previously collaborated with partners like AWS and Google through Project Glasswing to use Mythos Preview for discovering thousands of zero-day vulnerabilities, including a 27-year-old issue in OpenBSD, and employed a layered security classifier strategy: Fable 5 is public-facing, while Mythos 5 is limited to vetted defense parties. The Dario Amodei team has coordinated security assessments with the government multiple times before releasing cutting-edge models.
The Department of Commerce is mobilizing export control resources to implement precise controls on models with high cybersecurity reasoning capabilities, forcing Anthropic to shift from rapid commercial iteration to strict compliance and government cooperation, while accelerating resource allocation towards domestic infrastructure and sharing with allies to maintain technological leadership.
Similar to the impact of early semiconductor export controls on entities like Huawei, and the pre-release government review faced by companies like OpenAI, Anthropic is currently transitioning from independent cutting-edge R&D to a controlled framework under national security.
This essentially represents a capital concentration driven by regulatory changes: breakthroughs in AI network offense and defense capabilities that exceed physical and cryptographic limits have triggered a reassessment of sovereign risks. The government is reconstructing model distribution pathways through export controls, shifting from open commercial use to a licensing system, thereby strengthening the U.S.'s dominant position in global AI pricing power and security infrastructure.
ABAB News · Cognitive Law
The stronger the capability, the stricter the regulation; openness is momentarily pleasant, compliance is long-term beneficial.
Jailbreaking is not an accident, but a proof of capability; regulation must catch up with technology to balance risk leverage.
Layered filtering protects commercial use, while a complete freeze signifies sovereignty; technology knows no borders, but control has boundaries.