OpenAI Gov Briefings

OpenAI spent the past week briefing U.S. agencies, state governments and Five Eyes partners on a new cyber product as it prepares a tighter rollout for powerful security tools. (reuters.com) Axios reported the briefings included demonstrations of the product’s capabilities and its access controls, and one event in Washington drew about 50 cyber defense practitioners from government and allied groups. (axios.com) The product appears tied to OpenAI’s Trusted Access for Cyber program, which the company expanded on April 14 and used to introduce GPT-5.4-Cyber to vetted defenders and security teams. (openai.com) OpenAI said that program is being scaled to “thousands” of verified individual defenders and “hundreds” of teams, with identity checks and layered safeguards meant to keep stronger cyber features with approved users. (openai.com) Cyber models can help defenders scan code, spot weaknesses and speed incident response, but the same capabilities can also help attackers find and exploit flaws faster. OpenAI’s Preparedness Framework lists cybersecurity as a frontier-risk category alongside biological risks and artificial intelligence self-improvement. (openai.com) That framework says models with higher cyber capability require stronger controls before broader deployment. OpenAI’s developer documentation says GPT-5.3-Codex was the first model it classified as “High” in cybersecurity capability, triggering extra safeguards and monitoring. (cdn.openai.com) (developers.openai.com) OpenAI has also said it is training models to refuse clearly malicious requests, using automated systems to flag suspicious cyber activity, and routing high-risk cases for added review. (developers.openai.com) The government outreach comes as artificial intelligence labs compete to get their cyber tools in front of public-sector buyers and national-security officials. Reuters noted earlier this month that Anthropic launched “Project Glasswing,” a preview program that gives selected partners access to an unreleased model. (reuters.com) OpenAI said last week it is also working with the Center for AI Standards and Innovation and the United Kingdom AI Security Institute to evaluate the model’s cyber capabilities and safeguards before access widens further. (openai.com) For now, the message from the briefings is narrow: OpenAI is not treating advanced cyber assistance like a normal software launch, and governments are being shown the locks before the keys. (openai.com)