OpenAI and more than 100 technology firms have signed an open letter saying AI-powered cyberattacks are about to leave the lab and hit real-world infrastructure. The coalition warns that hackers will soon use large language models and autonomous agents to hunt for flaws and launch multi-stage exploits at machine speed, jeopardizing hospitals, power grids and water supplies.

Why the warning matters now

Artificial-intelligence models have sprinted from research curiosities to services that generate code, write emails and even design hardware in seconds. That productivity boost hands malicious actors tools that can scan networks, write exploit scripts and coordinate attacks without human hands. A joint alert from three U.S. agencies confirmed that attackers are already using AI to produce exploit code targeting industrial control systems such as the Siemens S7 series, which run many energy, water and manufacturing plants.

What’s driving the threat

The letter highlights three technical shifts that turn AI into a weapon. First, large language models can turn a brief description of a target into working code, slashing the time needed to build a payload. Second, reinforcement-learning agents can test thousands of exploit variants in a sandbox, learning which versions succeed. Third, generative tools can craft phishing messages that mimic a victim’s style, boosting credential-theft rates. Combined, a single automated pipeline can locate a vulnerability, produce a tailored exploit and fire it at dozens of devices in minutes.

Targets that could be hit

Critical infrastructure draws the focus because its systems blend digital control with physical processes. A compromised water-treatment plant could misdose chemicals; a hacked power substation could plunge a region into darkness; a tampered hospital network could disable life-support equipment. The letter calls out hospitals, water utilities, energy grids and chemical manufacturers as the most exposed sectors. Many of these operators run legacy equipment that cannot be patched quickly, leaving a huge attack surface for AI-driven scripts.

How defenders can keep pace

The signatories argue that the “defender’s edge” still exists but is shrinking. To hold onto it, they propose three actions:

  • Corporate accountability – Make security a boardroom priority, matching budgets and oversight to the scale of the AI threat.
  • Governmental support – Boost funding for research, share threat intelligence openly and coordinate response plans with private operators.
  • Democratizing security – Build affordable, easy-to-deploy AI-based defense tools so smaller utilities and municipal services aren’t left as weak links.

Potential pushback

Some experts caution that spotlighting AI as a new danger could distract from long-standing gaps such as unpatched software, weak passwords and poor network segmentation. They point out that many of the same AI techniques touted for offense can also defend—automated code review, anomaly detection and rapid incident response. The letter itself notes that defenders already use AI, but critics warn the industry may over-promise new tools while under-investing in basic hygiene.

What to watch

  • Policy initiatives – Expect proposals for dedicated funding streams for AI-focused cybersecurity research and standards governing the release of powerful generative models.
  • Tool releases – Vendors are likely to launch AI-augmented vulnerability scanners and intrusion-detection systems aimed at the sectors the letter highlights.
  • Incident reports – The first publicized AI-driven breach of a critical-infrastructure system would validate the coalition’s alarm and could trigger regulatory action.

The open letter marks the first time so many competitors have spoken with one voice about a shared existential risk. Whether the industry turns that consensus into concrete safeguards will decide if AI fuels the next wave of cyber devastation or becomes a tool that keeps the lights on.