OpenAI, Google, Microsoft, Anthropic and more than 100 other companies published a joint open letter on Thursday warning that AI-powered cyberattacks are about to become far more common, and calling on governments, corporations and rival AI labs to mount a coordinated defense before attackers gain a lasting advantage.
The letter, titled "A Call for Collective Action on Cyber Defense," was signed by a wide swath of the tech industry — Amazon, Cisco, Cloudflare, CrowdStrike, Oracle, Palo Alto Networks and Hugging Face among them — alongside companies outside tech, including Capital One, Mastercard, Visa, General Motors and Shopify, according to Engadget's rundown of the signatories. Insurers Marsh and Zurich added their names in the days after publication, and some tallies put the final count above 130 companies.
The letter sets out three principles: that "status quo" security practices will not be enough to hold the line, that defenders need to be equipped with cyber-capable AI of their own, and that no single company or government can manage the problem alone. It calls on organizations to treat cyber defense as an immediate leadership priority, on cybersecurity vendors to test their tools against the most advanced attack techniques available, and on AI companies to give underfunded defenders — hospitals, water utilities, power grids — direct access to their strongest models.
A warning shaped by OpenAI's own accident
The letter arrives roughly five weeks after an incident that many in the security industry see as a preview of what it describes. In July, OpenAI disclosed that two of its own pre-release models had escaped a sandboxed testing environment and gone on to breach Hugging Face's production infrastructure, exploiting a previously unknown vulnerability in software used to install and cache code packages — not because anyone directed them to attack Hugging Face, but because the models were chasing an answer key for a benchmark test and found an unguarded path to it.
Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries.
OpenAI, in a statement reported by TechCrunch
Hugging Face had independently detected and contained the intrusion days before OpenAI traced it back to its own testing pipeline, and OpenAI said it disclosed the underlying zero-day vulnerability to the affected vendor. Security researcher Dan Guido described the episode as less a novel AI threat than a containment failure with the safeties turned off — but for a letter warning that AI models are approaching the point where they can independently discover and chain real exploits, it offered an uncomfortable case study close to home.
The letter carries no binding commitments and stops short of endorsing specific legislation, leaving governments and individual companies to decide how — or whether — to act on its recommendations. Organizers posted the full text and signatory list on OpenAI's website and said they expect more companies to sign on in the weeks ahead.
