OpenAI's New AI Model Faces Unprecedented Government Scrutiny

Key points

• OpenAI limits GPT-5.6 Sol release to a small group approved by the Trump administration as of June 2023.
• Anthropic's AI models Fable 5 and Mythos 5 were pulled offline under a similar directive earlier this month.
• The U.S. government is reviewing AI models for up to 30 days under a new executive order signed by Trump.

OpenAI is holding back the wide release of its latest artificial intelligence model, GPT-5.6 Sol, amid concerns from the Trump administration that powerful AI could pose cybersecurity risks. The new model is currently available only to a select group of partners pre-approved by the administration.

OpenAI announced this limited access Friday, describing it as a temporary measure. The company hopes to make the model broadly available in the near future, but acknowledged the need for caution. The government’s recent actions follow similar restrictions imposed on GPT-5.6 Sol's competitor, Anthropic, which had to take down its AI models Fable 5 and Mythos 5.

Earlier in June, President Trump signed an executive order requiring AI developers to submit their most advanced systems for government review to assess national security threats. Although participation is technically voluntary, the framework to guide these reviews remains under construction.

Should customers be worried?

For now, customers have limited access to the new GPT-5.6 Sol model because the Trump administration is vetting its potential cybersecurity risks. This is part of a broader effort to ensure that AI technologies don't inadvertently aid criminals in finding software vulnerabilities.

OpenAI asserts that GPT-5.6 Sol is primarily designed to help identify and fix these vulnerabilities, rather than exploit them. However, the company also acknowledges unforeseen risks when combining its model with other tools, which is why it released the model in stages with additional safety measures.

The company has not disclosed the names of the approximately 20 customers currently using GPT-5.6 Sol. Meanwhile, cybersecurity experts, including Stanford University's Alex Stamos, have criticized the administration's tactics. Stamos argued that these actions might hinder the U.S. in AI innovation, as similar AI models are already available in other countries.

In an ongoing series of negotiations, OpenAI CEO Sam Altman has been in discussions with Commerce Secretary Howard Lutnick about the release strategy. Anthropic's CEO, Dario Amodei, has had a more contentious experience, with the Pentagon labeling the company a national security risk.

The government's concerns stem partly from Anthropic's claims that its Mythos model could be misused to identify software vulnerabilities, potentially threatening computer networks globally. Until a comprehensive review framework is established, AI companies like OpenAI and Anthropic will continue to face scrutiny over their developments.