OpenAI GPT-5.6 Sol release restricted after US government cybersecurity vetting request

OpenAI said it is limiting access to its new artificial intelligence model, GPT-5.6 Sol, after a request from President Donald Trump’s administration. The company said the model will go first to a small set of trusted partners. Those partners must be approved by the Trump administration. OpenAI said this review step is not ideal long term.

Representative image

The move is part of wider White House scrutiny of powerful AI systems. Officials said they are working with frontier AI labs on security issues. The focus is on models that might create cybersecurity dangers. The government has looked closely at tools that could help find software flaws. Such flaws can be used to attack critical computer networks.

OpenAI GPT-5.6 Sol release limits and phased rollout

OpenAI said the restricted rollout is temporary and it expects broader access in coming weeks. "We dont believe this kind of government access process should become the long-term default,\" OpenAI said. OpenAI also said Sol is stronger at helping users find and fix vulnerabilities. The company said Sol does not cross OpenAI’s internal risk threshold for harm.

OpenAI still flagged uncertainty around new uses and unexpected risks. It said risk can increase if Sol is combined with other tools. \"That uncertainty, along with the models broader step change in capabilities, is why we are pairing the models increased capabilities with stronger safeguards and a phased release,\" the company said Friday. OpenAI has not identified any of the approved customers.

Trump administration AI oversight and the 30-day vetting framework

Trump earlier in June signed an executive order on AI oversight. The order set up a plan for the federal government to review national security risks. It allowed up to 30 days of review before public release. The order described developer participation as voluntary. The full framework has not yet been completed.

OpenAI’s decision followed earlier government action involving Anthropic. Anthropic is the maker of the Claude chatbot and is also based in San Francisco. Anthropic took offline two new models, Fable 5 and Mythos 5. The company did so days after launching them. It said it acted to meet a Trump directive blocking use by foreign nationals.

Anthropic Claude models, cybersecurity concerns, and government actions

Officials became more concerned after Anthropic warned about Mythos earlier this year. Anthropic said Mythos was skilled at spotting software weaknesses. The company said those abilities could be weaponised by malicious hackers. Such use could threaten key computer networks worldwide. The White House said it continues working with AI labs on these scaling risks.

Some Trump allies blamed Anthropic and CEO Dario Amodei for increased scrutiny. \"Dario came to Washington a few months ago, back in April, and basically said that he had created a cyber weapon called Mythos,\" said investor David Sacks, on a recent podcast. \"And he spiked the cortisol level, got everyone really worried.\"

Criticism of OpenAI GPT-5.6 Sol access decisions and US policy

US Rep Lori Trahan, a Massachusetts Democrat, criticised how access decisions are being made. Trahan is a co-author of a bipartisan AI regulation bill. Trahan said the administration is choosing model access company by company. \"No law. No process. No oversight. Just appointees in Washington deciding whos in and whos out.\"

Technology experts also criticised actions that led to Anthropic shutting down Fable. Anthropic had described Fable as a safer version of Mythos. It has been unavailable for two weeks. \"I just want to say that pretty much nobody in the cybersecurity industry believes that theres any factual basis for this action,\" said Alex Stamos.

Stamos is a Stanford University cybersecurity expert and Corridors chief product officer. Stamos is also a former chief security officer at Facebook parent Meta. Stamos said Amazon reviewed research on Fable, then shared analysis. Stamos said the review did not show risks beyond other public models. It included models made in China.

\"If the administration is honest about wanting the United States to beat China in this race, then this is about the dumbest thing they could possibly do,\" Stamos said. The debate has raised questions about how policy choices affect US firms. Critics said unpredictable government moves may slow releases. They also said it can weaken competitiveness.

OpenAI CEO Sam Altman spoke with US Commerce Secretary Howard Lutnick on Wednesday. The talk covered the model release and came after weeks of negotiations. Other AI leaders have also met Trump officials. Anthropic took part too, though Amodei has had a more contentious relationship with the administration. Those talks continued as access controls expanded.

The Pentagon designated Anthropic as a national security risk over ethical and safety concerns. Trump also ordered federal agencies to stop using Claude. Anthropic responded with a lawsuit that is still moving through federal courts. OpenAI, meanwhile, said its staged Sol release is meant to manage risk. The company said wider availability is expected soon.

With inputs from PTI