OpenAI and Anthropic have restricted access to their latest AI models, GPT-5.6 Sol and Mythos 5, to U.S. government-approved customers amid a cybersecurity review led by the Trump administration, according to reports from the Associated Press and The Guardian. The move follows an executive order signed by President Donald Trump in June 2026 that mandates federal oversight of advanced AI systems for 30 days before public release.
Companies Cite Government Pressure, Warn of Long-Term Risks
OpenAI stated it is limiting access to its GPT-5.6 Sol model to “trusted partners” after the Trump administration requested a staggered rollout, a decision the company described as a “temporary step” toward broader availability. “We don’t believe this kind of government access process should become the long-term default,” the company said in a statement. The Guardian reported that OpenAI’s CEO, Sam Altman, confirmed the government would approve access on a “customer-by-customer” basis during the preview period, with a wider release planned “a couple of weeks later” if the process proceeds smoothly.
Anthropic, OpenAI’s rival, faced similar restrictions earlier this year when the U.S. Commerce Department blocked its Mythos 5 model from foreign nationals. The company said the Trump administration lifted those restrictions Friday, allowing the model to be “redeployed to a small group of cyber defenders and infrastructure providers.” However, Anthropic’s CEO, Dario Amodei, has criticized the government’s approach, with investor David Sacks, a Trump adviser, claiming Amodei “created a cyber weapon called Mythos” during a visit to Washington in April. “He spiked the cortisol level, got everyone really worried,” Sacks said on a podcast, though he acknowledged the model’s “advanced cyber capabilities” had some basis in reality.
Both companies emphasized their models are designed to “find and fix vulnerabilities” rather than execute cyberattacks. OpenAI’s GPT-5.6 Sol, the strongest in its series, is “better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks,” the company said. However, OpenAI acknowledged “unforeseen risks” if the model is combined with other tools, prompting its phased release strategy.
Government Scrutiny Intensifies Amid AI Advancements
The Trump administration’s actions reflect growing concerns about the dual-use potential of advanced AI systems. Officials have raised alarms since Anthropic warned in early 2026 that its Mythos model could identify software flaws with “weaponizable” capabilities, posing threats to critical infrastructure. The executive order issued in June 2026 established a framework for federal agencies to evaluate national security risks of “frontier AI systems” before their public debut, though the process remains untested.
The White House has maintained collaborative ties with AI developers, stating it continues to “engage with frontier AI labs on addressing the challenges of scaling the fast-growing technology.” However, critics argue the process lacks transparency. “The Trump administration is deciding company by company who gets access to the newest AI model. No law. No process. No oversight. Just appointees in Washington deciding who’s in and who’s out,” a cybersecurity expert told The Guardian.
Anthropic’s phased release of Mythos 5—initially delayed voluntarily before being mandated by the government—mirrors OpenAI’s approach. Both companies are navigating a delicate balance between innovation and security, with OpenAI’s GPT-5.6 series including three versions: Sol (strongest), Terra (mid-tier), and Luna (lowest-cost). While Sol is restricted to U.S.-based entities, OpenAI plans to expand access to “supported countries” like the UK and Australia in the coming weeks.
Industry Reactions and Uncertain Future
The cybersecurity community remains divided. While some experts support stricter controls, others warn of stifling progress. “That uncertainty, along with the model’s broader step change in capabilities, is why we are pairing the model’s increased capabilities with stronger safeguards and a phased release,” OpenAI said, echoing concerns about unintended consequences. However, critics argue the government’s involvement risks creating a “two-tiered” system where only select entities benefit from cutting-edge AI tools.
“Users, developers, enterprises, cyber defenders, and global partners who need them” are being excluded, OpenAI’s statement noted, highlighting the tension between security and accessibility. The Guardian reported that some in the industry doubt the long-term viability of the government’s approach. “Pretty much nobody in the cybersecurity industry believes there’s any factual basis for this action,” a source said, though the statement was not attributed to a specific organization.
Looking ahead, the next 30 days will be critical. The executive order’s framework for vetting AI systems remains under development, and the outcomes of OpenAI and Anthropic’s restricted releases could set precedents for future AI regulation. As the White House continues its review, the broader implications for innovation, security, and global competition in AI remain unclear.
For more on this story, visit AP News and The Guardian.
Find more reporting in our Business section.
