OpenAI and Anthropic Restrict New AI Models During Trump Administration Cybersecurity Review
OpenAI and Anthropic are limiting access to their newest artificial intelligence models as President Donald Trump’s administration expands its cybersecurity review of frontier AI systems.
OpenAI said Friday that its new GPT-5.6 series will begin with a restricted rollout to selected trusted partners rather than a full public release. The model family includes GPT-5.6 Sol, its flagship system, along with Terra and Luna, which are positioned as lower-cost options. OpenAI said Sol is its most capable model to date, with stronger performance across coding, biology and cybersecurity tasks, and that the preview group’s participation has been shared with the U.S. government.
The company said it plans to make the models more widely available in the coming weeks, including through ChatGPT, Codex and the API. During the preview period, access will be limited to a small group of partners and organisations, with OpenAI framing the approach as a temporary step while it continues work with federal officials on cyber safeguards for advanced models.
The Associated Press reported that GPT-5.6 Sol will initially be available only to customers approved through the Trump administration’s review process. The report said OpenAI has not publicly named the roughly 20 customers cleared to use the model so far.
Anthropic is facing similar scrutiny. Earlier this month, the Claude maker said the U.S. government had issued an export control directive requiring it to suspend access to Fable 5 and Mythos 5 for foreign nationals. Anthropic said it disabled access for all customers in order to comply with the order.
Fable 5 had been introduced as Anthropic’s most capable widely released model, while Mythos 5 was designed for a smaller group of cyber defenders and infrastructure providers through Project Glasswing. Anthropic described Mythos 5 as a high-powered cybersecurity model with some safeguards lifted for approved defensive use.
The AP reported Friday that restrictions on Mythos 5 had been eased, allowing it to return to a limited group of cyber defenders and infrastructure providers. Fable 5, however, remained caught up in the wider access dispute.
The restrictions follow a June 2 executive order that directed federal agencies to build a process for evaluating advanced AI models with cyber capabilities. The order calls for a voluntary framework under which developers can give the government access to covered frontier models for up to 30 days before release to other trusted partners, while also collaborating on which early users should receive access.
The developments mark a new phase in the relationship between Washington and leading AI labs. Supporters of tighter review argue that frontier models capable of advanced software engineering and vulnerability discovery may create risks if misused. Critics warn that unclear government involvement in customer selection could slow innovation, introduce political pressure and leave companies uncertain about how future model launches will be handled.
For now, both OpenAI and Anthropic are moving ahead cautiously, offering their most advanced systems to narrow groups while they negotiate the boundaries of AI safety, national security and commercial deployment.
Photo Credit: DepositPhotos.com