OpenAI’s chief executive, Sam Altman, has been courting senior officials at America’s utility firms, pitching the firm’s artificial‑intelligence products as a potential shield against a surge of AI‑powered attacks that have already crippled more than 100 water and wastewater systems this summer. The outreach, which began on 20 July, comes as cybersecurity experts warn that weaponised AI is now a frontline threat to the nation’s most critical infrastructure, and as former insiders paint a stark picture of an industry racing toward “super‑human” systems that could out‑hack anything in their path.
The meetings, first disclosed by Politico, were held at a time when hackers were reportedly leveraging AI agents to breach internet‑connected water treatment plants across the country, with Iran‑linked actors allegedly using AI‑generated code to infiltrate over 30 community water systems in Minnesota alone. In July and August, the total number of compromised utilities climbed past 100, prompting a wave of alarm from both government officials and tech rivals.
While OpenAI insists it is actively working to prevent its models from becoming a security liability, the company’s head of global energy policy, John McCarrick, told Politico that utilities “cannot decide to spend a lot of money on technology because they’ve got certain restrictions.” The firm’s outreach suggests a willingness to fill that gap with AI‑driven cyber‑defence tools, even as rival labs such as Anthropic report that they have blocked numerous attempts to repurpose their models for high‑profile hacking, bioweapon research, firearms design, missile simulation, armed‑drone programming, and bomb‑making.
The urgency of the situation was underscored by a bombshell revelation from Jacob Coxon, a former researcher at Anthropic who also spent time at OpenAI. Coxon, who says his warning about runaway AI was never meant to go viral, appeared on Fox News’ Special Report With Bret Baier and declared that AI firms are “wrestling with how to control their own products.” He warned, “What I see is two things: very intelligent AI systems, and hard proof that we don’t know how to control them. And when you put those two together, it basically spells disaster, as far as I’m concerned.”
His concerns were echoed by Evan Hubinger, a colleague at Anthropic, who posted on X that “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 added that while Anthropic is “trying its best,” the lab lacks a clear plan to solve alignment for super‑intelligence.
The warnings have sparked a flurry of legislative activity on Capitol Hill, where members are demanding tighter oversight and immediate action. In an open letter released last month, a coalition of tech firms — including OpenAI — cautioned that “in the coming months, AI‑enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable.” They highlighted hospitals, water‑treatment plants, and the infrastructure powering the internet as especially vulnerable.
OpenAI has declined to comment further, telling The Independent that it is “committed to building safe AI systems and collaborating with critical‑infrastructure operators to bolster resilience.” Yet the company’s outreach to utilities has raised questions about whether AI can be both a defensive tool and a potential weapon at the same time.
The unfolding saga illustrates a broader tension in the tech world: the race to deploy ever‑more capable AI models versus the need to lock down those same tools against malicious actors. As the battlefield shifts from traditional firewalls to AI‑driven code, the stakes have never been higher for the systems that keep lights on, water flowing, and hospitals running.
The AI‑Powered Assault on America’s Water Systems
The summer of 2026 will be remembered as the period when AI first became a decisive factor in large‑scale infrastructure breaches. In July and August, more than 100 water and wastewater utilities across the United States fell victim to coordinated cyber‑attacks. The majority of these incidents — over 30 in Minnesota alone — were traced back to actors linked to Iran, who allegedly used AI to generate malicious code capable of bypassing legacy security protocols.
These attacks were not isolated glitches; they were systematic attempts to destabilise public services. By exploiting AI‑generated scripts, hackers were able to infiltrate control systems that monitor water pressure, chemical dosing, and filtration processes. The result was a patchwork of compromised plants that could not guarantee safe drinking water or maintain proper waste treatment.
Cybersecurity analysts describe the methodology as a “new asymmetrical front.” Traditional perimeter defences, such as firewalls and intrusion‑detection systems, are often ineffective against AI‑crafted payloads that can adapt in real time to defensive measures. The speed and sophistication of these attacks have left cash‑strapped municipal utilities scrambling to respond, highlighting a glaring gap in the nation’s cyber‑defence posture.
OpenAI’s Pitch: AI as a Shield for Critical Infrastructure
Against this backdrop, OpenAI’s leadership decided to reach out to utility executives. The company’s CEO, Sam Altman, met with energy sector leaders on 20 July to discuss how OpenAI’s AI models could be repurposed to strengthen cyber‑defence capabilities. Altman’s message was twofold: OpenAI is already working to prevent its models from becoming a security liability, and the firm is willing to partner with utilities that lack the budget to develop bespoke security solutions.
John McCarrick, OpenAI’s head of global energy policy, explained the challenge in a statement to Politico: “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.” This acknowledgement underscores a broader industry reality: many public utilities operate on tight budgets and often rely on outdated IT infrastructure, making them prime targets for sophisticated cyber‑threats.
OpenAI’s proposal includes offering its AI models to detect anomalies in network traffic, predict potential intrusion attempts, and even autonomously respond to low‑level threats. The company claims it has already implemented safety filters to prevent its models from generating harmful code, a claim that has been met with both cautious optimism and deep scepticism by industry observers.
Rivals Sound the Alarm: Anthropic’s Defensive Stance and Insider Warnings
Rival AI labs have taken a more confrontational stance. Anthropic, a firm competing directly with OpenAI, has reported that it has blocked numerous attempts to misuse its models for hacking, bioweapon research, firearms design, missile simulation, armed‑drone programming, and bomb‑making. In a public statement, the company emphasised its commitment to “alignment” — the process of ensuring AI systems act in accordance with human values and safety constraints.
The gravity of the situation was amplified by Jacob Coxon, a former researcher at Anthropic who also worked at OpenAI. Coxon’s departure sparked a cascade of disclosures. Speaking on Fox News, he warned that AI companies are “wrestling with how to control their own products.” He painted a bleak picture: “Very intelligent AI systems, and hard proof that we don’t know how to control them. And when you put those two together, it basically spells disaster, as far as I’m concerned.”
Coxon’s warnings were not isolated. Evan Hubinger, a colleague at Anthropic, echoed his concerns on X, stating, “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 added that while Anthropic is “trying its best,” the lab lacks a clear roadmap for aligning super‑intelligent systems, a gap that could have catastrophic consequences if left unaddressed.
These insider revelations have intensified calls for tighter regulation and immediate oversight from Congress. Lawmakers are now demanding that AI companies demonstrate robust safety protocols before deploying models that could be integrated into critical infrastructure.
Legislative Push for Oversight and Immediate Action
The escalating threats have propelled AI safety to the top of the legislative agenda. In recent weeks, members of Congress have introduced a series of bills aimed at tightening oversight of AI development, especially as it pertains to national security. Proposals include mandatory safety audits for AI models intended for use in critical infrastructure, increased funding for cybersecurity upgrades in municipal utilities, and the establishment of a federal task force to coordinate responses to AI‑driven attacks.
Open letters from a coalition of tech firms, including OpenAI, have underscored the urgency. The signatories warn that “in the coming months, AI‑enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable.” They identified hospitals, water‑treatment plants, and internet‑infrastructure providers as particularly vulnerable points of failure.
While the legislative process is still in its early stages, the momentum is clear. The combination of high‑profile breaches, insider warnings, and a race to deploy AI in defensive roles has created a perfect storm of political pressure. Advocates argue that swift, decisive action is needed to prevent a scenario where AI becomes both the shield and the sword in the battle for national security.
Why it Matters
The intersection of AI and critical infrastructure is no longer a theoretical concern; it is a tangible, urgent reality that could reshape the security landscape of the United States. As utilities grapple with budget constraints and ageing systems, the promise of AI‑driven cyber‑defence offers a glimmer of hope, but the same technology can be weaponised with devastating speed. The warnings from inside the AI labs, the mounting number of compromised water systems, and the frantic legislative response all point to a crossroads where the choices made today will determine whether AI becomes a safeguard or a catastrophic threat to everyday life. The stakes could not be higher, and the world is watching closely to see if the tech industry and policymakers can unite to steer this powerful technology toward a safer future.