In an unprecedented move, the Trump administration on Friday, June 13, 2024, ordered Anthropic PBC to immediately disable all foreign access to its advanced AI models, Fable 5 and Mythos 5, citing national security concerns. The US government is imposing these surprise export controls on advanced AI models, but these same models have already demonstrated vulnerabilities to 'jailbreaks' shortly after release. This rapid pace of AI development and its inherent security challenges will likely lead to more aggressive and immediate government interventions, potentially stifling global collaboration and accelerating a balkanization of AI capabilities.
The Unprecedented AI Export Ban and Its Rationale
The Trump administration's directive on June 13, 2024, ordered Anthropic to disable foreign access to its Fable 5 and Mythos 5 AI models, a move confirmed by Bloomberg, The New York Times, and Al Jazeera. The Commerce Department cited national security concerns, forcing the company to pull its latest release (Fox Business). This comprehensive directive targets specific advanced models, yet its efficacy is immediately questioned by a critical vulnerability: Amazon AI experts 'jailbroke' Fable 5 after its June 9, 2024 release, exposing its full cyber capabilities (Fox Business). This occurred before the government's export controls were even implemented.
The 'unprecedented' nature of these controls, imposed on models already proven vulnerable, suggests a reactive government approach to AI security threats. Such broad measures, enacted post-vulnerability, overlook inherent technological weaknesses. The immediate disabling of foreign access to Mythos 5 and Fable 5 implies a high-stakes threat perception, but Fable 5's 'jailbreak' indicates the government may be attempting to secure systems that are fundamentally porous. This raises critical questions about the efficacy of such controls when inherent vulnerabilities persist, potentially offering only superficial security.
The rapid, reactive nature of these controls, coupled with the demonstrated vulnerabilities of advanced AI models, suggests a future where regulatory interventions become both more frequent and less predictable. If AI development continues its current trajectory, global collaboration in this critical field will likely fragment further, creating a complex and potentially less secure landscape for technological advancement.
