US restricts public GPT-5.6 Sol release

The government-mandated restriction of GPT-5.6 Sol marks a significant escalation in state oversight of frontier AI models.

• –**Precedent of Intervention**: This follows a similar export control restriction placed on Anthropic's "Mythos 5" earlier in the month, demonstrating a systematic shift toward active government vetting of frontier LLMs.
• –**Dual-Use Cyber Capabilities**: Sol's primary trigger for restriction is its unprecedented capacity to find and remediate software vulnerabilities, which safety teams worry could be reverse-engineered for autonomous cyberattacks.
• –**OpenAI's Pushback**: OpenAI is complying but explicitly warned against making government-vetted distribution the default, arguing it stifles legitimate defensive research and developer innovation.
• –**The Developer Impact**: Individual developers and non-approved enterprises are left in the dark, unable to use OpenAI's most capable reasoning and agentic model for their own systems.
• –**Rollout Timeline**: Although restricted to roughly 20 approved partners for now, OpenAI plans to coordinate with the White House to expand access to a wider preview in the coming weeks.