The United States government has partially lifted its restrictions on Anthropic's powerful Mythos 5 artificial intelligence model, clearing the company to release the system to a select group of vetted US organizations. The decision, first reported on June 26, 2026, reverses an earlier suspension that had sidelined one of the most capable AI models in the country over acute cybersecurity concerns.
Bloomberg, Reuters, the New York Times, and the Wall Street Journal all confirmed the rollback, marking a pivotal moment in the escalating debate over who gets access to frontier AI. The move also reshapes the competitive landscape at a time when US firms are racing to maintain a lead over Chinese rivals in the broader AI industry. For more context on this story, see our ongoing AI trends.
A Model That Broke Into Classified Systems
Mythos 5, part of Anthropic's Claude lineup alongside the companion Claude Fable 5 model, was suspended by the US government in mid-June 2026 after it demonstrated an unsettling capability: the ability to find and exploit vulnerabilities in classified American government systems. According to reporting by the Associated Press and Reuters on June 24, the model surfaced security flaws in sensitive federal infrastructure, triggering an immediate halt to its deployment.
The concerns were severe enough that Security Affairs reported the model had penetrated "almost all" of the National Security Agency's classified systems within hours during testing. BBC covered the original suspension on June 13, noting that both Mythos 5 and Claude Fable 5 were pulled from circulation "over security fears." The Conversation later explained that the government intervention was driven by the system's demonstrated offensive-security potential rather than conventional safety worries about bias or harmful text.
What the Rollback Actually Allows
The new arrangement does not restore Mythos 5 to general availability. Instead, the Trump administration has partially lifted the export ban to permit a controlled re-release. Reuters reported that Anthropic can now deploy Mythos to "trusted" US organizations, while Semafor described the recipients as "some US companies." CNBC specified that the permitted users include "some companies, government agencies," and the Associated Press framed the broader trend as new AI models being limited to "Trump-approved customers" during an ongoing cybersecurity review.
The New York Times characterized the shift as the US "loosening restrictions" on the model, and the Wall Street Journal reported that the administration "rolls back part" of the earlier ban rather than rescinding it entirely. Politico's headline described it bluntly: the Trump administration "partially lifts Anthropic's AI export ban." NBC News reported that the government gave Anthropic the "green light for limited re-release of Mythos 5," while The Hill confirmed that federal authorities permitted release "to select companies."
In short, Mythos 5 is back, but only for a vetted few.
Part of a Broader Government Clampdown
The Mythos decision does not exist in isolation. It arrives alongside a parallel restriction on OpenAI, which agreed to stagger the release of its new GPT-5.6 models and limit them to trusted partners at the government's request. The Information first reported that request on June 25, and TechCrunch noted that OpenAI itself said the restrictions "shouldn't be the template" going forward. WIRED summed up the public-facing reality bluntly: "OpenAI Has New AI Models. Here's Why You Can't Use Them."
Together, the moves signal that the US government is asserting a more direct role in deciding who can access the most powerful AI systems, framing advanced models as sensitive technology akin to controlled exports. That posture carries significant implications for national security, commercial competition, and the open-source community.
The Geopolitical Stakes
The restrictions and their partial relaxation also play into the global AI race. CNBC reported that China's Zhipu is "closing in on top US AI models" precisely as American leaders like Anthropic and OpenAI are being held back by security reviews. With domestic models temporarily constrained, the competitive gap with Chinese labs could narrow, a tension that policymakers are visibly weighing as they calibrate how much to restrict and how much to release.
Why It Matters
The Mythos episode crystallizes a dilemma that will only sharpen as models grow more capable. On one hand, an AI system that can autonomously discover vulnerabilities in classified networks is exactly the kind of capability that demands careful stewardship and government oversight. On the other hand, locking frontier models behind government approval risks concentrating power, slowing commercial innovation, and ceding ground to international competitors operating under no such constraints.
For Anthropic, the partial green light is a qualified win: the company can begin rebuilding momentum around its flagship model, but only within tight guardrails. For the broader industry, it establishes a precedent that the most advanced AI systems may now require what amounts to a government license to deploy, a framework closer to pharmaceutical regulation or export control than the open, fast-moving ecosystem AI developers have operated in until now.
As the cybersecurity review continues and additional models come up for approval, expect the line between "trusted" and "restricted" to remain the defining fault line in American AI policy.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
