Key Takeaways
- OpenAI has temporarily limited the release of its new AI model, GPT‑5.6 Sol, to a small group of trusted partners approved by the Trump administration.
- The move follows a similar voluntary shutdown of Anthropic’s Fable 5 and Mythos 5 models after a White House directive blocking their use by foreign nationals.
- President Trump’s June executive order established a framework for up‑to‑30‑day federal vetting of advanced AI systems for national‑security risks, though participation is nominally voluntary.
- Government officials cite concerns that powerful models could be weaponized to find and exploit software vulnerabilities, threatening critical infrastructure.
- Critics, including cybersecurity expert Alex Stamos, argue there is little factual basis for the restrictions and warn they could handicap U.S. competitiveness against China.
- OpenAI CEO Sam Altman has engaged in negotiations with Commerce Secretary Howard Lutnick, while Anthropic’s Dario Amodei faces a more contentious relationship, including a Pentagon designation as a national‑security risk and an ongoing lawsuit.
- OpenAI stresses that Sol is better at helping users find and fix flaws than at conducting cyberattacks, but acknowledges unforeseen risks when combined with other tools.
- The company describes the current restricted rollout as a temporary step toward broader availability in the coming weeks.
Government Request Prompts Restricted Release
OpenAI announced on Friday that it is limiting the release of its newest artificial‑intelligence product, GPT‑5.6 Sol, to a “small group of trusted partners” approved by the Trump administration. The decision comes at the explicit request of White House officials, who have expressed unease about the model’s potential cybersecurity implications. OpenAI characterized the move as a precautionary measure rather than a permanent policy shift, emphasizing that the restricted availability is intended to be temporary while the company works toward broader distribution in the near future.
OpenAI’s Stance on Temporary Vetting
In its statement, OpenAI warned that the current government‑access process should not become the long‑term default for AI releases. The firm described the testing period as a temporary step on the “path to broader availability in the coming weeks,” suggesting that it views the administration’s involvement as a short‑term safeguard. By framing the restriction as a voluntary, interim measure, OpenAI seeks to balance compliance with federal concerns while preserving its commitment to rapid innovation and widespread access to its technology.
Parallel Actions Against Anthropic
The OpenAI restriction mirrors earlier actions taken against its rival Anthropic, which voluntarily took offline two newly released models—Fable 5 and Mythos 5—just days after making them publicly available. Anthropic complied with a Trump administration directive that barred the use of its models by foreign nationals, citing national‑security worries. Both incidents illustrate a pattern of heightened government scrutiny targeting the most capable generative‑AI systems emerging from U.S. labs.
White House Collaboration and Executive Order
The White House said it continues to work with frontier AI labs to address the challenges posed by the fast‑scaling technology. Earlier in June, President Trump signed an executive order on AI oversight that creates a framework for federal agencies to vet the national‑security risks of the most advanced AI systems for up to 30 days before public release. Although the order describes participation by developers as voluntary, the administration has been applying pressure that effectively makes compliance a prerequisite for deployment.
Concerns Over Cyber Weapon Capabilities
Officials have grown increasingly anxious after Anthropic warned that its Mythos model was particularly advanced iterations of its model could identify software flaws with a precision that could be weaponized by malicious hackers, jeopardizing critical computer networks worldwide. The administration fears that similar capabilities in OpenAI’s Sol—or any future model—could be exploited to compromise infrastructure, prompting the pre‑emptive vetting process.
Industry Criticism of Government Intervention
A broad coalition of cybersecurity experts has denounced the government’s actions, arguing that they lack a solid evidentiary foundation. Stanford University’s Alex Stamos, formerly chief security officer at Meta and now a product officer at AI security firm Corridor, stated that reviews of Anthropic’s Fable model—conducted with Amazon’s cloud‑computing team—revealed no risks beyond those present in other publicly available AI models, including those developed in China. Stamos warned that if the administration truly wants the United States to outpace China in AI, imposing such restrictions is counterproductive.
Expert Opinions on Risk Assessment
Stamos further noted that the prevailing sentiment within the cybersecurity community is that the current government vetting does not reflect a genuine threat landscape. He argued that the administration’s actions appear to be driven more by perception than by concrete technical analysis, potentially stifling innovation without delivering proportional security benefits. Other experts echo this view, cautioning that overly cautious policies could push cutting‑edge research overseas or into less‑transparent channels.
Executive Dialogues and Legal Battles
OpenAI CEO Sam Altman has been engaged in a series of negotiations with U.S. Commerce Secretary Howard Lutnick regarding the Sol model’s release, part of broader talks between AI industry leaders and Trump administration officials. Anthropic’s CEO Dario Amodei, meanwhile, has experienced a more strained relationship: the Pentagon labeled Anthropic a national‑security risk for raising ethical and safety concerns about AI use in warfare, and Trump himself directed federal agencies to cease using the Claude chatbot. Anthropic responded with a lawsuit that remains pending in federal court, highlighting the growing friction between innovators and policymakers.
Outlook for Broader Availability
Despite the current restrictions, OpenAI maintains that Sol is primarily a defensive tool—better at helping users locate and fix vulnerabilities than at executing cyberattacks—and that it does not cross the company’s internal risk threshold. Nevertheless, the firm acknowledges that unforeseen dangers could arise if the model is combined with other utilities, justifying the phased release and enhanced safeguards. OpenAI expects that, after the temporary vetting period concludes, the model will become more widely accessible within the coming weeks, pending further evaluation.
Conclusion and Implications
The episode underscores the escalating tension between rapid AI advancement and governmental efforts to mitigate perceived national‑security risks. While officials stress the need to prevent potential cyber weaponization, industry voices warn that pre‑emptive restrictions may undermine U.S. competitiveness and impede innovation. As the executive order’s framework matures and more models undergo scrutiny, the balance between security safeguards and open technological progress will remain a critical point of debate for policymakers, corporations, and the broader public.