US Emergency Directive Forces Sweeping AI Model Shutdown

Last week, the US government sent a shockwave through the AI industry by issuing an emergency national security directive that effectively forced Anthropic to pull its most advanced Fable 5 and Mythos 5 AI models globally. The move was unprecedented in scope, affecting not only US-based users but all international access to these flagship systems. Companies and developers reliant on Anthropic's API for mission-critical applications found themselves suddenly locked out, with no prior warning or detailed explanation from authorities.

The directive came under the purview of US export control laws, which have increasingly been used to restrict the spread of dual-use technologies—tools that can serve both beneficial civilian purposes and pose significant national security risks. In this case, the focus was on Mythos 5, a model described by Anthropic as its most powerful AI system to date. Unlike earlier models optimized for general chat and content generation, Mythos 5 was specifically designed for cybersecurity operations: it can scan, identify, and fix severe vulnerabilities in complex, supposedly unbreakable codebases. However, the same capability can be reversed by malicious actors to discover and exploit critical infrastructure flaws. This duality made Mythos 5 a prime candidate for stringent government oversight.

To manage this risk, Anthropic had already implemented a restricted access program called 'Project Glasswing.' Under this initiative, only about 150 trusted entities were granted access to Mythos 5. The partner list included major technology firms such as Google, NVIDIA, Microsoft, and Apple. The goal was to create a controlled environment where the model's power could be used to harden national cybersecurity without proliferating dangerous capabilities. Yet the US government's intervention suggests that even this limited distribution was not sufficient to allay security concerns.

Anthropic's Optimistic Outlook for Restoration

Speaking at a press conference in Seoul, Mr. Chris Ciauri, Anthropic's Managing Director of International, stated that the company is 'very confident that in the coming days, the models will become available again,' as reported by Korea JoongAng Daily. The event was originally intended to mark Anthropic's expansion into the South Korean market, but the urgent questions about the export controls and Project Glasswing dominated the discussion. Mr. Ciauri declined to comment on most questions related to the specific reasons for the shutdown, indicating that sensitive negotiations with government agencies were still ongoing.

Anthopic's optimism is grounded in ongoing technical and diplomatic efforts to satisfy Washington's requirements. The company has been working on implementing additional guardrails, security patches, and possibly geo-fencing to ensure that access to Mythos 5 and Fable 5 can be reinstated without exposing the models to unwarranted risk. The precise nature of these new controls has not been disclosed, but they likely involve more granular access logs, enhanced monitoring of API usage patterns, and possibly restrictions on the type of queries that can be processed in certain regions.

The Core Dispute: SK Telecom's Alleged Ties to China

According to a report by Wired, the US government's decision to impose export controls on Anthropic's models materialized after a dispute over including South Korean telecom giant SK Telecom in Project Glasswing and granting it access to Claude Mythos 5. Citing 'people familiar with the matter,' the report notes that US officials were concerned about what they alleged were SK Telecom's ties to China. These allegations stem from SK Telecom's existing business partnerships in the telecommunications and digital services sectors, some of which involve Chinese companies. US intelligence agencies reportedly flagged SK Telecom as a potential vector for technology transfer to China, despite the company's strong ties to South Korea and its membership in the US-led security alliance.

SK Telecom has vigorously denied these claims. An SK Telecom representative told South Korea's The Chosun Daily, 'The anonymous insider's remarks in foreign media lack verified facts, and our company has no ties to China.' The company pointed out that it adheres to all international export control regulations and has a long history of cooperation with US firms, including partnerships with major tech companies for joint development of AI and 5G technologies. However, in the current geopolitical climate, where the US and China are competing for technological dominance, even the perception of a security risk can trigger immediate action.

Background: The Evolution of US AI Export Controls

The incident with Anthropic is part of a broader trend by the US government to tighten controls on advanced AI and semiconductor technologies. Similar measures have been applied to companies like NVIDIA, which was restricted from exporting its high-end GPUs to China without a license. The rationale is to prevent US-origin technology from being used to enhance the military capabilities of potential adversaries. In the case of AI models, the challenge is even greater because software is intangible and can be copied or transferred digitally. This has led to calls for 'digital export controls' that can be enforced through cloud platforms and API endpoints.

Anthropic itself was founded by former members of OpenAI who split off to focus on AI safety and constitutional AI. The company has always positioned itself as a responsible actor in the AI space, voluntarily implementing safety measures like those seen in Project Glasswing. Nevertheless, the federal government's decision to intervene indicates that self-regulation is not always seen as sufficient by national security agencies.

Implications for the AI Industry and Global Partnerships

The freezing of Anthropic's models has sent ripples across the AI ecosystem. Developers who built applications on top of Fable 5 or Mythos 5 were forced to switch to less powerful alternatives, potentially delaying product launches and research projects. The incident also raises questions about the viability of international AI partnerships. Countries like South Korea, which have strong technological sectors but are geographically close to China, may find themselves caught in the crossfire of US-China tensions. For Anthropic, the goal is to demonstrate that it can maintain strict control over its most advanced systems while expanding its global footprint.

Mr. Ciauri's confidence that the models will be restored soon reflects a belief that a compromise can be reached. However, it is unclear whether the US government will impose permanent restrictions on certain geographic access. Some analysts predict that the outcome could involve a tiered system where users in allied nations like South Korea retain full access, while other regions face limitations. For now, the focus is on restoring services to paying customers who have been locked out, and the next few days will be critical in determining how this high-stakes dispute is resolved.

Source: Android Authority News