GPT-5.6 Sol Access Limited After US Cybersecurity Review

OpenAI has significantly limited access to its new artificial intelligence model, GPT-5.6 Sol. This restriction follows directives from the Trump administration. Only a select group of trusted partners, who have undergone a government cybersecurity assessment, can access the model. OpenAI states this is a transitional phase, anticipating a broader rollout within weeks. The company emphasizes this type of government access review should not become the long-term standard. The White House is working with AI labs on challenges related to scaling AI. President Trump recently issued an executive order establishing a protocol to evaluate potential national security threats from advanced AI systems. This order allows for review periods of up to 30 days. OpenAI says the Sol model is designed to identify and fix software vulnerabilities, not to execute cyberattacks. The company acknowledges potential unexpected risks, especially when Sol is integrated with other tools. This uncertainty, combined with the model's enhanced capabilities, necessitates robust safeguards and a gradual release. This cautious approach follows Anthropic's withdrawal of two Claude models, Fable 5 and Mythos 5, after a Trump directive barred their use by foreign nationals. Anthropic had previously warned about Mythos's potential to exploit software vulnerabilities. This ongoing situation highlights the evolving intersection of advanced AI development and national security concerns.