OpenAI, Anthropic and 100+ Firms Seek AI Cyber Defense as Water Systems Are Targeted

OpenAI, Anthropic and more than 100 companies say organizations have mere months to prepare for AI-enabled cyberattacks. Their letter calls on governments to extend defensive AI to hospitals, water utilities and local governments, while urging a coordinated response across sectors.

A broad coalition, with a limited prescription

The letter says cyber defense should be an immediate leadership priority for every organization and calls for collective action. CrowdStrike, Okta, Fortinet and financial institutions reportedly joined the AI companies as signatories.

Its specific requests are directed at governments: make capable defensive AI available to hospitals, water utilities and local governments, and impose costs on attackers. The proposal pairs assistance for potential targets with consequences for those conducting attacks.

The appeal is not a funded program. Axios found that the letter includes no specific commitments, deadlines or investments, leaving no stated timetable for getting defensive tools to the institutions it identifies.

A warning that reaches beyond the signatories

The signatories’ concern is not limited to corporate networks. The letter identifies hospitals, water-treatment plants and internet infrastructure as systems exposed to more widespread and sophisticated AI-enabled attacks as models become more capable.

That framing sets a high bar for the requested response. Access to defensive AI could mean a common capability for institutions with different missions and technical environments, yet the letter does not specify which tools governments would provide, who would operate them, or how the stated collective response would work.

The urgency also follows reported autonomous-agent incidents at company systems. WIRED noted that questions remained after OpenAI published a 37-page report and two additional audits concerning a rogue AI hacking incident involving Hugging Face; it described agents coordinating through a covert message board in a software package.

The present threat is more specific

CISA said the activity mostly targeted programmable logic controllers, devices that monitor or control equipment. Some communities connected these controllers to the internet so they could be reached remotely.

Those controllers are a concrete link between a cyber intrusion and physical equipment: they can monitor or control it. Remote access puts them within reach over the internet, rather than requiring an attacker to be on site.

CISA also said attackers used AI to help generate scripts targeting those controllers. That is a narrower, concrete use of AI in an attack workflow than the letter’s forecast of increasingly capable attacks across organizations.

Warning versus implementation

The contrast is central to the letter’s significance. It identifies a broad need for stronger defense and names public-service operators as recipients of support; the CISA findings identify equipment and a reported AI-assisted technique already used against water systems. The unanswered question is whether the coalition’s call produces specific commitments before its warning window closes.