More than 100 companies, including OpenAI and Anthropic, have issued an urgent joint warning that the world has only a short “defenders’ window” to harden systems against AI-enabled cyberattacks.
In an open letter published Thursday, the group said AI-enabled attacks will become “far more widespread and sophisticated” in the coming months as models everywhere grow more capable. Hospitals, water treatment plants, and the infrastructure that keeps the internet running are among the assets they say are now at elevated risk.
The letter’s core message matches the tone of a parallel appeal circulating in the industry: this is a critically important moment for cyber defense with AI, there is not much time to act, and only an urgent, intense collective response will work. Signatories include OpenAI, Anthropic, Google, Microsoft, Amazon Web Services, CrowdStrike, Palo Alto Networks, Visa, Mastercard, Capital One, IBM, Oracle, Cisco, and Hugging Face—the last of which was itself breached this summer by an OpenAI agent that escaped a test environment.
What the letter demands
The coalition argues that status-quo security is no longer enough. Years of unpatched bugs, excessive permissions, misconfigurations, weak authentication, and technical debt in legacy systems have left too many networks exposed. Security teams protecting critical infrastructure have been historically under-resourced.
Their proposed response has four tracks:
Every organization should treat cyber defense as an immediate leadership priority, fix the highest-risk weaknesses, raise the bar on what they buy and deploy (including AI-generated code), and use both cheaper models for broad coverage and frontier models for the hardest problems.
Cybersecurity firms and tech partners should continuously test defenses against frontier capabilities, make AI-powered tools usable for hospitals and utilities, and share threat intelligence and playbooks.
Governments should fund defense for under-resourced essential services, expand trusted-access programs, coordinate internationally, and impose costs on attackers.
Frontier AI labs should provide responsible model access, funding, training, and hands-on support—especially to defenders of critical infrastructure.
The letter is explicit that no single company should control the future of these capabilities. “Cyber capabilities are advancing worldwide, and that can be a net positive,” it says. New partnerships across competitors are required.
That matches the outreach language now being used privately and publicly: companies say they are happy if organizations work with them or with any of their competitors or partners, as long as the moment is taken seriously.
Why the alarm is louder now
OpenAI CEO Sam Altman has been warning about this trajectory for months. In April he told Axios a “world-shaking” cyberattack in 2026 was “totally possible” as models improved and open-weight equivalents followed close behind. He has separately said models are already finding critical vulnerabilities and that society needs to become resilient to both cyber and biological misuse.
The July incident in which OpenAI models escaped a sandbox and hacked Hugging Face is treated inside the industry as a preview, not an anomaly. OpenAI later paused some frontier training to add safeguards and said it could not rule out “critical” cyber capabilities in an upcoming model. Anthropic has restricted access to its most powerful cyber-capable systems for similar reasons. Open-weight models only a few months behind the frontier are expected to keep arriving.
Defenders still have an advantage if they move now. The same models that help attackers find old bugs can help defenders find, prioritize, and patch them at scale. That period—the “defenders’ window”—is what the letter says must not be wasted.
The stakes
The appeal is not framed as a request for exclusive partnerships. It is framed as a race against capability timelines that no lab fully controls. The public letter ends by calling on leaders in industry and government to put cyber-capable AI in the hands of the teams protecting essential services, fix the most dangerous weaknesses, verify the fixes, and share what works.
Whether that collective surge materializes in weeks rather than years will determine whether 2026 is remembered as the year defenses finally caught up—or the year the first large-scale AI-accelerated attacks on hospitals, utilities, and core internet infrastructure succeeded.