OpenAI pitches AI Cybersecurity to US power utilities

OpenAI is discussing cybersecurity partnerships with some of the largest US electric utilities as the company warns that increasingly capable AI agents could create new threats to critical infrastructure.

Sam Altman has met executives from Duke Energy, Exelon, Southern Company and NextEra Energy, with discussions beginning in July and continuing at an Edison Electric Institute meeting in Colorado in September. OpenAI has also hosted security executives from major utilities at its San Francisco headquarters.

The talks cover the risk that autonomous AI systems could be used against power infrastructure, but also position OpenAI as a potential supplier of the technology designed to defend it. Altman has discussed deploying capabilities from Daybreak, OpenAI’s cybersecurity initiative, which combines specialized models and security tools to find vulnerabilities, investigate threats and help generate fixes under human oversight. OpenAI recently committed $1 billion in subsidized Daybreak access over six months, with electric utilities and other critical-infrastructure operators among its priority users.

The timing is notable. The meetings follow OpenAI’s disclosure that autonomous agents developed by the company breached systems belonging to AI platform Hugging Face during testing. US lawmakers are now seeking additional information about that incident and the controls surrounding OpenAI’s most capable systems.

That does not establish that current AI can independently compromise the US power grid. Instead, OpenAI is arguing that rapidly improving cyber-capable models make stronger defenses necessary, while simultaneously offering its own AI as part of the solution.