Startup AI Models Frozen: OpenAI, Anthropic Restrict Access Under Trump's 30-Day Review

On June 26, 2026, OpenAI and Anthropic publicly revealed that they were restricting access to their latest artificial intelligence models, GPT-5.6 Sol and Mythos 5 respectively, following cybersecurity reviews conducted by the Trump administration. This marks an unprecedented level of direct U.S. government scrutiny over private-sector AI releases, moving beyond voluntary guidelines toward what is effectively a gatekeeping role. OpenAI announced it would limit Sol to a small group of customers vetted by the administration, while Anthropic secured approval for a limited redeployment of Mythos 5 to cyber defenders and infrastructure providers—a partial reversal of a de facto ban imposed two weeks earlier. Both companies framed the moves as temporary, with OpenAI stating it does not believe such government access processes should become the long-term default. The actions flow directly from a June 2026 executive order signed by President Trump that establishes a framework for federal agencies to assess advanced AI systems for up to 30 days before their public release on national security grounds. Although the order is described as voluntary, the rapid compliance by two of the largest AI labs suggests the sector sees little room to refuse participation without risking regulatory or reputational fallout. The catalyst for this heightened oversight was Anthropic's earlier warning that its Mythos model possessed formidable capabilities for discovering software vulnerabilities—tools that could be weaponized by malicious hackers to compromise critical infrastructure. That dual-use dilemma forced a reckoning: the same technology that could supercharge cyber defense could also enable devastating cyberattacks. The administration's response, pairing restriction with selective release to trusted partners, attempts to thread that needle but raises fundamental questions about who decides which models are too dangerous and for whom. Industry implications are profound. For the first time, a U.S. administration has effectively delayed and shaped the rollout of frontier AI models, setting a precedent that could permanently alter the innovation lifecycle. AI companies may now need to build government review timelines into product roadmaps, potentially slowing the breakneck pace of release that has characterized the sector. International competitors, especially in China and the EU, will be watching closely: the U.S. is now signaling that national security can override open competition in AI, which may provoke reciprocal barriers abroad and fragment the global AI ecosystem. For domestic startups and researchers, the chilling effect could be acute, as restricted access to state-of-the-art models widens the gap between a few well-funded labs and the broader innovation community. Meanwhile, the cybersecurity community gains insight into how powerful vulnerability-detection AI might be deployed responsibly, but also faces new ethical lines around using tools that governments have deemed too dangerous for unrestricted access. Looking ahead, the temporary nature of the current restrictions may be tested if the review framework becomes entrenched. The 30-day review window could expand, and the definition of “trusted partners” could narrow. The AI industry, long accustomed to self-regulation and rapid iteration, now confronts a future where government clearance becomes a prerequisite for deployment of the most capable systems. How courts may respond to legal challenges over prior restraint or free speech issues remains an open question, but for now, the Trump administration has established a powerful new lever of control over the technologies shaping the 21st century.