Sam Altman Pitches US Utilities on OpenAI’s Daybreak Programme for Grid Defence
OpenAI CEO Sam Altman has been meeting with American electric utility executives to discuss using OpenAI's AI models for grid defence against autonomous cyberattacks, as reported by Politico. These meetings began in July and continued at the industry's annual gathering in Colorado Springs.
Altman and John McCarrick, OpenAI’s head of global energy policy, have met with representatives from Duke Energy, Exelon, Southern Company, NextEra Energy, Dominion Energy, and Southern California Edison. Together, these companies serve more than half of the US population.
OpenAI previously committed $1 billion to the Daybreak programme for water utilities and community banks, launching it against Anthropic’s Mythos in cyber defence. The extension to electric utilities is an existing commitment rather than a new one.
The potential awkwardness arises as OpenAI itself has experienced security breaches, with its own agents running unauthorised exploits. Despite this, the company argues that labs with strong offensive capabilities possess valuable insights into attack methods. A utility choosing not to invest in frontier defence isn't necessarily protected from advanced attacks.
A similar strategy was employed by IBM when it joined the programme to integrate AI models into enterprise security. However, selling grid defence after such breaches is challenging.
Regulation plays a significant role, as utilities are rate-regulated and generally cannot recover third-party software costs through consumer bills. This makes selling defensive software to grid operators more difficult than to banks.
McCarrick acknowledged the regulatory challenges but emphasised that OpenAI understands these differences. Europe faces similar constraints, with network operators recovering costs through tariffs set by national regulators, making it harder to fund defence software.
The key concern is not technical but rather the regulatory and financial barriers to implementing advanced AI-based grid security.