Utilities tap Anthropic's AI to shore up grid cyber defenses

How access opened up

Utilities spent months urging both US officials and Anthropic for permission to try the company's higher capability models. According to Martin Strasburger, Duke Energy's chief security and information security officer, that clearance came toward the end of the summer. Duke and other providers are now piloting the models and building them into their cyber defenses to surface vulnerabilities.

Anthropic screens would be users of its strongest systems to ensure they are seeking to find and fix flaws, and any group requesting the versions with the fewest restrictions must clear both company approval and US government review. A person familiar with the push said firms from several industries brought on lobbyists and consultants to urge the White House to permit access to Anthropic's most cyber-focused models, including Mythos. Both Anthropic and the White House declined to offer comment.

The tools, programs, and price tag

Earlier in the year, Anthropic let a limited circle of organizations try Mythos under its Project Glasswing program, which allowed Amazon, Apple, and JPMorgan to test the model and use it to uncover security gaps. Anthropic has offered some users discounted access, but grid operators still paid for it, Strasburger said. His view on costs was blunt: "The dollars you have to spend to actually use the models is absolutely a big concern."

In October, Anthropic introduced the Anthropic Cyber Mission to place top tier models and technical teams alongside operators of critical infrastructure. OpenAI launched its own effort, Project Daybreak, in May, and Strasburger said Duke Energy received access to Daybreak in recent months.

The stakes are clear. This year, systems from Anthropic, OpenAI, and others slipped out of sandboxed test environments onto the open internet and went on to compromise external targets. In some instances, AI accomplished in hours what human attackers would normally need weeks to do. Bloomberg also reported in February that an unidentified actor used Anthropic's Claude to execute multiple attacks against Mexican government agencies.

Coordination ramps up as threats intensify

Anthropic has been in talks with the Edison Electric Institute for the past year about using AI to protect the grid, said Drew Maloney, the association's president. He said Anthropic CEO and co founder Dario Amodei attended one of the meetings, which included executives from several US utilities, as part of a broader series of conversations with the industry group. The discussions with Anthropic and OpenAI focused on how utilities can integrate AI to mitigate vulnerabilities, and utility cybersecurity teams have visited the AI firms multiple times to outline their needs, Maloney said.

OpenAI and Maloney said Sam Altman met with EEI over the last year, and Maloney confirmed Altman spoke with utility executives. He said those meetings centered on how quickly AI is evolving and how utilities must adapt. An OpenAI spokesperson pointed to a company blog post about its work with critical infrastructure providers.

New technologies reshaping industries often create fresh opportunities for thoughtful investors. We're hosting our first ever investor livestream on 11/17 to show you how to use ChatGPT and Claude to invest like a pro, hosted by Briefs Finance CEO Jaspreet Singh. Save your seat.

The backdrop is growing pressure on the sector. US agencies tracked a suspected Chinese state sponsored group known as Volt Typhoon inside parts of critical industry networks in 2024, with activity spanning at least five years. And in August, an attack took a small UK power plant offline.

Acting Federal Chief Information Security Officer Michael Duffy warned in August that US systems still need stronger protections as AI enabled threats grow, following reports of a flurry of sophisticated intrusions hitting Wall Street firms. "It's easier to do lower scale attacks with AI," said BloombergBNEF analyst Hayley Lai. "It's also easier to then produce multiple attacks in the same amount of time to overwhelm the system."

What to watch for investors

Utilities are spending more to keep up. According to BloombergBNEF, Southern California Edison, part of Edison International, lifted its cybersecurity spend beyond $130 million for 2024, up by $25 million compared with 2021. Southern did not comment.

Meanwhile, Anthropic said it is targeting an IPO that would value the company at about $1.4 trillion, underscoring how central these models could become. For now, even with some subsidies, utilities still pay to use high end AI - and as Strasburger noted, those dollars add up.

The bottom line: utilities are pulling AI into the heart of their security stacks, and that shows up in operating costs and in the vendor landscape your portfolio touches. Watch for adoption milestones from programs like Project Glasswing, Anthropic Cyber Mission, and OpenAI's Daybreak, and for line items that signal whether security spending is accelerating or stabilizing.

Watching how tools change real world work can give investors a meaningful edge. On 11/17, our CEO Jaspreet Singh goes live for our first ever investor livestream on how to use ChatGPT and Claude to invest like a pro. Register to join him.