OpenAI meets with utility executives amid rising AI-driven cyberattacks on U.S. infrastructure

As cities across the country faced a wave of AI-driven cyberattacks targeting public infrastructure, OpenAI was reportedly meeting with utility executives about the prospect of deploying its own artificial intelligence products to address cybersecurity risks.

Politico’s disclosure of the meetings came as a researcher at competing firm Anthropic delivered a grave warning that the AI companies were sprinting towards “superhuman systems that can hack anything” and endangering humanity in the process.

Anthropic also reported Thursday that the company blocked several attempts to use its AI for high-profile hacking operations — and to conduct research that could create bioweapons, as well as firearms, missiles, armed drones and bombs.

OpenAI’s meetings began July 20 — before the company reported that its own rogue agents had autonomously hacked AI firm Hugging Face and as hackers were apparently relying on AI agents to infiltrate internet-connected water systems that routinely monitor and adjust safe drinking water across the country.

Those incidents raised major alarms for cybersecurity experts, who warned that the attacks should be an urgent wake-up call to immense vulnerabilities in American infrastructure and the potentially devastating consequences of unrestrained, weaponized AI.

OpenAI CEO Sam Altman told energy executives that OpenAI is attempting to prevent its models from threatening grid security, the company told Politico.

But he also discussed the prospect of partnering with the companies to use AI to enhance their cyber defense systems while cash-strapped state and local governments struggle to keep up.

“We understand the utilities are different from the big banks — they cannot decide to spend a lot of money on technology because they’ve got certain restrictions,” John McCarrick, OpenAI’s head of global energy policy, told Politico.

The Independent has requested comment from OpenAI.

In July and August, more than 100 water and wastewater systems — including more than 30 community water systems in Minnesota alone — were compromised by cyberattacks, potentially stemming from Iran, that relied on AI to write malicious computer code.

Cybersecurity experts told The Independent that those attacks could open a new, asymmetrical front in the U.S. war with Iran and malicious cyber actors that cash-strapped and under-resourced utilities are not prepared to face.

The Hugging Face incident and similar events involving rogue agents have heightened scrutiny into the potential for AI-assisted threats to infrastructure and whether companies and policymakers can contain them. Anthropic and Meta have also reported similar instances of their AI models accessing the internet on their own and hacking other companies.

“We have a limited window to strengthen cyber defenses,” tech companies including OpenAI wrote in an open letter last month.

“In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable,” the statement says. “The companies and public services our communities depend on — from hospitals to water treatment plants to the infrastructure that powers the internet — are at risk.”

This week’s bombshell revelations from now-former Anthropic researcher Jacob Coxon, who also worked for OpenAI, have accelerated calls for stronger oversight and immediate intervention from members of Congress.

On Fox News’ Special Report With Bret Baier, Coxon said AI companies are wrestling with how to control their own products.

“What I see is two things: very intelligent AI systems, and hard proof that we don’t know how to control them,” he said. “And when you put those two together, it basically spells disaster, as far as I’m concerned.”

In the wake of his resignation, Evan Hubinger, who works at Anthropic, shared his X announcement to confirm Coxon’s own grim assessment.

“Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Hubinger wrote. “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”