OpenAI's New AI Model Faces Unprecedented Government Scrutiny

Key points

• OpenAI has restricted GPT-5.6 Sol to roughly 20 Trump-approved partners as of June 2026.
• Anthropic's Fable 5 and Mythos 5 were pulled offline under a separate Trump directive blocking foreign-national access.
• A Trump executive order signed in June allows up to 30 days of government vetting for advanced AI systems.
• Participation is described as voluntary, but the review framework is not yet fully built.

OpenAI said Friday it is restricting its new AI model, GPT-5.6 Sol, to a "small group of trusted partners" approved by the Trump administration, citing cybersecurity concerns about powerful AI systems. The company called it a temporary step toward "broader availability in the coming weeks."

The move follows the White House's earlier action against rival Anthropic. As we reported on 13 June, Fable 5 and Mythos 5 went dark after a Trump directive blocked foreign nationals from accessing them. Fable has now been unavailable for two weeks.

The pressure on Anthropic started when CEO Dario Amodei told Washington officials in April that Mythos had advanced enough cyber capabilities to find software flaws in ways that could threaten critical networks. Investor David Sacks, who co-leads Trump's science and technology advisory council, described that briefing plainly on a recent podcast: "He basically said that he had created a cyber weapon called Mythos. And he spiked the cortisol level, got everyone really worried."

Trump signed an executive order in June requiring developers to submit advanced AI systems for up to 30 days of national-security review before public release. Participation is voluntary, but the rules governing those reviews have not been written yet.

Should customers be worried?

For now, GPT-5.6 Sol is only accessible to the roughly 20 approved customers OpenAI has not publicly named. The core concern is whether a powerful vulnerability-finding model could be turned against the systems it was meant to protect.

OpenAI says Sol "is better at helping people find and fix vulnerabilities" than at enabling attacks, and that it does not breach the company's own internal risk threshold. It also acknowledged that combining the model with other tools could produce unforeseen risks, which is why it paired the rollout with stronger safeguards.

That framing has not satisfied critics. Stanford cybersecurity expert Alex Stamos, chief product officer at AI security firm Corridor and a former chief security officer at Meta, told reporters this week that he reviewed Amazon's analysis of Fable and found no risks beyond those already present in publicly available AI models, including ones made in China. "If the administration is honest about wanting the United States to beat China in this race," he said, "then this is about the dumbest thing they could possibly do."

OpenAI CEO Sam Altman spoke with Commerce Secretary Howard Lutnick on Wednesday about the release. Anthropic has also been in talks with the administration, but from a harder position: the Pentagon has designated the company a national security risk, Trump ordered federal agencies to stop using Claude, and Anthropic's lawsuit challenging those actions is still in federal court.

The real watch item here is the review framework itself. Until it exists, "voluntary" means whatever the White House decides it means that week.