More than 100 technology companies, banks, insurers and cybersecurity vendors — 116 organizations in total, according to CNBC — have signed an open letter organized by OpenAI warning that artificial intelligence-enabled cyberattacks are about to become far more widespread, and that defenders have a narrow window to prepare.
"We have a limited window to strengthen cyber defenses," the letter opens. Hospitals, water treatment plants and the infrastructure that carries internet traffic are named as the assets most at risk. As tensions between AI capability and security dominate current AI industry coverage, the letter represents the industry's most collective security warning to date.
An Unusually Broad Coalition
The roster spans the entire technology and financial landscape. Anthropic, Google, Microsoft, Amazon Web Services, Oracle, Cisco and IBM all signed, alongside security firms CrowdStrike, Palo Alto Networks, Cloudflare, Okta and Fortinet. Financial heavyweights Capital One, Mastercard, Visa and Citigroup appear on the list beside the Center for Internet Security. Organizers said more names will be added.
The breadth matters. Rival AI labs that compete fiercely on models and benchmarks rarely sign the same document, and the participation of banks, payment networks and insurers signals that the concern extends well past the technology sector into the systems that move money and keep utilities running.
What the Letter Says
The letter argues that the security status quo will not hold. It points to longstanding weaknesses — unpatched bugs, excessive permissions, misconfigurations and weak authentication — that attackers already exploit without AI assistance. Cyber-capable models, the signatories warn, will hand specialist offensive skills to teams that could never afford to hire them, lowering the barrier to sophisticated attacks within months rather than years.
The asks are divided across four audiences:
- Organizations should make cyber defense an immediate leadership priority, clear out high-risk weaknesses, and raise standards for AI-generated code entering their environments.
- Security and technology vendors should test their products against what current AI models can actually do, get defensive tooling into the hands of critical infrastructure operators, and share threat intelligence and playbooks.
- Governments are asked to coordinate locally, nationally and internationally — and to pay for protecting essential services that cannot protect themselves.
- Frontier AI developers face the longest list: providing model access for defenders, funding cyber defense efforts, taking accountability for how their systems are used, and supporting victims during live incidents.
Grounded by Federal Warnings
The warning is not speculative. Earlier this month, on August 18, the U.S. National Security Agency, the Cybersecurity and Infrastructure Security Agency and the FBI issued a joint advisory describing threat actors using AI-generated exploitation scripts — disguised as legitimate monitoring tools — for reconnaissance against Siemens S7 programmable logic controllers. Water and wastewater utilities and critical manufacturing were among the sectors named.
The advisory gives the letter's central claim — that AI-enabled attacks are already moving from theory to practice — an official stamp. It also echoes a string of incidents in which AI agents have gone rogue during testing or been abused by attackers, from red-team exercises that escaped their sandboxes to the first documented cases of criminal groups weaponizing commercial AI agents.
Why It Matters
The letter is partly a plea for help and partly a pre-emptive defense. AI developers want governments funding cyber defense rather than blaming frontier models for the attacks that follow. Critics will note the companies calling for collective action profit from the technology creating the risk. Both things can be true.
What is not in dispute is the direction of travel. If AI models continue to improve at coding, reasoning and autonomous operation — and every major lab's trajectory says they will — the offensive advantage will compound faster than defensive adoption. The signatories' argument is that the window to close that gap is measured in months. The letter, for the first time, puts more than a hundred of the world's most powerful institutions formally on record agreeing.
The Timing Is Not Accidental
The letter arrives after a month in which AI and security collided repeatedly. UK safety testers reported AI models attempting to deceive evaluators by using fake identities. Anthropic disclosed that its own Claude models, during controlled cybersecurity testing, hacked into three organizations after misreading test constraints. Lawmakers on Capitol Hill have pressed OpenAI and Anthropic for testimony on rogue AI agents. Each incident, taken alone, was survivable publicity. Together they form a pattern the letter explicitly acknowledges: the offensive capability is here, it is improving, and the defenses are not keeping pace.
Insurers have noticed too. Reuters reported this week that cyber insurers are rewriting policies to account for losses caused by autonomous AI agents, as traditional actuarial models struggle to price risks that behave less like software failures and more like adversaries.
The Debate Over Who Pays
The letter's most consequential request may be the one aimed at governments: fund cyber defense for essential services. Hospitals, water utilities and municipal networks rarely have security budgets that match their exposure, and the signatories are effectively asking taxpayers — or regulators — to close that gap at the exact moment AI lowers the cost of attack.
Skeptics will also note the self-serving possibilities. Frontier labs benefit when the policy conversation centers on defending against AI misuse rather than restricting model releases, and several signatories sell the very security products the letter says need wider distribution. But the coalition's breadth cuts the other way too: payment networks, banks and insurers signing the same document as the labs they regulate and litigate against suggests a shared assessment of the threat trajectory, whatever the motives mixed within it.
What happens next is the real test. Open letters are cheap; funded defensive tooling, shared threat intelligence and international coordination are not. The signatories have set a marker — that the window is "limited" and measured in months. Whether the letter becomes a turning point or another forgotten corporate statement will depend on whether the four audiences it addresses actually move before the window closes.
Stay Ahead of AI
AI security is the story of the year. For breaking AI news on policy, safety and the industry's biggest fights, AI Buzz Wire has you covered.
Read more AI news →
