The US government has made another significant reversal on Anthropic's Claude Mythos 5 AI model, this time removing export control restrictions that were imposed less than three weeks ago. According to reports from The Commerce Department, Commerce Secretary Howard Lutnick issued a letter to Anthropic on Tuesday lifting the export control directive that had banned non-US nationals from accessing the model since June 12. This latest development represents a major change in the Trump administration's approach to controlling frontier AI capabilities, coming after recurring flip-flopping on the issue. The reversal follows Anthropic's suspension of access to Mythos 5 for all users on June 13, after the initial ban included foreign employees at US companies and cyber defenders from international partners. The June 26 letter authorized partial restoration of Claude Mythos 5 to more than 100 US companies and federal agencies operating and defending critical infrastructure, marking the first restoration signal since the June 12 ban.
The Commerce Department's latest move allows Anthropic to coordinate with the government to expand Mythos 5 access to a broader group including international partners. As of July 1, Fable 5 – which Anthropic says has stronger safeguards than Mythos 5 – is available to public users globally. This represents a significant shift from the previous approach announced on June 26, when the US government had allowed Anthropic to release Mythos 5 but reserved access to only a select group of 'trusted' US companies and agencies. The administration's turbulent relationship with Anthropic has global consequences, including for US allies, with the EU previously losing access to Mythos 5 following the export control directive before presumably regaining it. The Annex A framework mirrors Anthropic's existing Project Glasswing structure, covering covered organizations, their foreign-national employees, and Anthropic's own foreign staff, with the letter explicitly stating that Anthropic has committed to working with the government on protocols, standards, and future releases.
The purpose of Project Glasswing is to help defenders get ahead of a coming shift in cybersecurity, with Anthropic and roughly 50 partners having already found more than 10,000 high- or critical-severity vulnerabilities across systemically important software. According to recent reports, Glasswing partners use the model for vulnerability scanning, patch writing, pre-release security checks, penetration testing and migrating legacy code into memory-safe languages. The project represents an early test of how powerful AI can be used defensively in cybersecurity before similar capabilities become widely available to attackers, with its promise being faster discovery of hidden flaws. However, the challenge lies in whether companies can fix these flaws quickly enough, as their software risk often sits not only in their own code but also in vendors, cloud providers and open-source components.
India remains actively pursuing access to Anthropic's Project Glasswing to enhance its critical digital infrastructure testing capabilities. According to reports from NDTV Profit, MeitY Secretary S Krishnan confirmed that the country is in talks with US authorities to secure access to this advanced AI platform. Speaking at the Confederation of Indian Industry Cybersecurity Summit, Krishnan emphasized that stress-testing India's systems on Anthropic's systems would be particularly useful for evaluating resilience against advanced AI capabilities. As Krishnan stated, "We're talking with American counterparts to get access for Project Glasswing," highlighting the importance of this collaboration for India's cybersecurity preparedness. The second Pax Silica Summit, hosted by the US State Department on June 25-26, 2026, expanded the US-led AI supply chain coalition from 25 to 35 nations, with new signatories including the European Union, Germany, the Netherlands, Argentina, Chile, Costa Rica, El Salvador, Greece, Kazakhstan, and Panama. India's formal request for assurances at the summit is the most diplomatically significant development, with Krishnan raising the kill switch concern directly: the Fable 5 ban had demonstrated that a single US government letter could cut off allied nations' access to frontier AI without prior consultation. India received an informal assurance, with Under Secretary for Economic Affairs Jacob Helberg describing India as a potential "comprehensive partner" under the initiative.
The US government's latest reversal comes as part of a broader framework established by President Trump's executive order this month, which creates a voluntary system for AI developers to offer "covered frontier models" to the US government for up to 30 days before releasing them to trusted partners. According to Kate Koren, an analyst at the Center for Strategic and International Studies, "The longer there isn't a system in place that will allow U.S. companies to widely release new models, the more likely it is that China will be able to catch up." The administration's approach has its flaws, with Chatham House experts previously arguing that tightening restrictions around valuable technology will not fully prevent their proliferation. This turbulent relationship between US political leaders and powerful technology companies highlights the unstraightforward nature of government control over AI development and deployment. Both Anthropic and OpenAI have now publicly committed to pre-briefing the government before future frontier model releases, with the August 1, 2026 deadline for NSA, Treasury, and CISA to build a classified frontier model benchmarking process and the July 8 deadline for Anthropic's ID verification rollout being the two structural dates that will shape how the governance regime operationalizes.
