Anthropic, the AI company behind the chatbot Claude, has reintroduced its advanced AI models Mythos and Fable after a US government ban citing national security risks. The ban was initially imposed due to concerns that these models could be exploited to identify vulnerabilities in critical infrastructure, potentially aiding hostile military intelligence from countries like China or Russia.
The US Commerce Department had enforced an export ban restricting foreign access to these AI tools, which Anthropic said was effectively impossible to enforce without a total ban, as verifying user nationality proved unfeasible. However, recent developments show the ban has been lifted, with Anthropic collaborating closely with the government to implement stronger safeguards and control mechanisms.
Restricted Access and Enhanced Safeguards
Mythos 5, designed specifically to detect cybersecurity vulnerabilities, is currently available only to selected trusted US organisations under the Glasswing program. Anthropic is working to broaden access to domestic and international partners within this initiative. Meanwhile, Fable 5, aimed at the general public, will be released with more robust safety measures.
Following reports from Amazon researchers who discovered ways to bypass Fable 5’s safeguards to exploit software vulnerabilities, Anthropic has introduced new protective layers. Any requests attempting such bypasses are now redirected to a safer model, Opus 4.8, prioritising security over user convenience.
Ongoing Challenges and Industry Collaboration
Despite these improvements, Anthropic acknowledges the difficulty of making AI models completely resistant to jailbreaks—techniques that circumvent safety controls. The company is actively engaging with the US government and partners like Amazon, Microsoft, and Google to establish common standards for detecting and mitigating such vulnerabilities.
This cooperative approach reflects a growing trend in AI governance, as the US administration also recently introduced a voluntary framework requiring developers to share new frontier AI models with the government prior to wider release. Rival OpenAI has faced similar restrictions, delaying the public launch of GPT-5.6 to comply with government requests.
Broader Implications and Future Outlook
Anthropic’s complex relationship with the US government highlights the tensions between innovation, security, and ethical use of AI. The Pentagon’s earlier designation of Anthropic as a "supply-chain risk"—due to the company’s refusal to support mass surveillance or autonomous weapons—illustrates these challenges.
As AI technologies evolve rapidly, the balance between accessibility and safeguarding national security remains delicate. Anthropic’s experience underscores the importance of collaborative frameworks and transparent protocols to ensure AI benefits while mitigating risks.
For businesses and developers interested in AI safety and governance, exploring advanced AI management tools is essential. JASON AI offers solutions to help organisations implement robust AI controls and maintain compliance with evolving standards. Learn more at https://jasonjuul.com.
Scope and Implementation Disclaimer: This article provides an overview of recent developments in AI security and government regulation based on publicly available information. It does not offer technical instructions or guarantee the efficacy of specific AI safety measures. Organisations should consult specialised security experts when implementing AI governance protocols.
For JASON AI product education, AI service categories, and current storefront information, visit https://jasonjuul.com.