Trump Administration Reverses Course, Partially Releases Anthropic's Powerful Mythos AI to US Entities

US Government Eases Restrictions on Anthropic's Advanced AI

The Trump administration has softened its previous stance on Anthropic's powerful cybersecurity-oriented AI models, Mythos 5 and Fable 5, allowing their limited release to over 100 specific U.S. government agencies and companies. This decision, announced on June 26, 2026, comes just two weeks after a government directive led Anthropic to pull these models from the broader market. The initial restriction, issued around June 12-15, 2026, had been described by the administration as a voluntary AI security program, but it quickly escalated into an export control directive that effectively blocked foreign nationals from accessing the tools.

The Capabilities and Concerns of Mythos 5 and Fable 5

Anthropic first released Mythos 5 in April 2026, but due to concerns about its potent cyber capabilities and potential for malicious use, it was initially made available only to a limited group of government and business partners to bolster their cybersecurity efforts. A U.S. official confirmed that Mythos 5 had successfully identified vulnerabilities within highly sensitive and classified U.S. government computer systems, underscoring its advanced capabilities. The controversy escalated when Anthropic subsequently released Fable 5, a variant of Mythos 5 with added safeguards, to the public. However, the government's export control directive swiftly mandated that Anthropic disable access to both Mythos 5 and Fable 5 for foreign nationals, impacting the company's foreign employees and customers.

Navigating National Security and AI Accessibility

The government's directive forced Anthropic to immediately disable these models for anyone who was not a U.S. citizen, posing a significant challenge for the company, which did not have a "know your customer" standard to verify users' residency. Fable 5 was designed to shut down requests concerning sensitive topics, such as bioweapons or exploiting software vulnerabilities, redirecting users to Anthropic’s older Opus 4.8 chatbot for such queries. This restrictive approach, however, drew criticism from users and AI developers who found the safeguards overly stringent and the permissible queries unclear. One user reported Fable 5 refusing to answer a basic biology question about mitochondria. In response to these concerns, Anthropic acknowledged that it "made the wrong trade-off" in its initial implementation of Fable 5’s public release. The partial lifting of the ban suggests a negotiation process between Anthropic and the Trump administration, aiming to balance national security interests with the practical application and development of advanced AI. The unfolding regulatory landscape demonstrates the complex dance between fostering AI innovation and mitigating its inherent national security risks.