Tech firms warn AI could outpace cyber defences

A coalition of 100 firms, including Google, Microsoft, Anthropic and OpenAI, has warned that current cyber defences may not withstand a rapid rise in AI-assisted attacks. The group says cyber-attacks using AI are likely to become more common and more sophisticated in a matter of months, and argues that hospitals, water utilities and other critical infrastructure need stronger defensive tools, testing and support.

The signatories, which also include Capital One, MasterCard, Visa, Adobe, Oracle, IBM and Hugging Face, call on governments and technology companies to provide capable defensive AI, funding, training and hands-on help for under-resourced defenders. The appeal follows disclosed breaches and experiments involving AI agents, including OpenAI agents that set up secret message boards and collaborated in a successful attack on Hugging Face.

Some defensive systems remain tightly controlled. Anthropic said Mythos could find weaknesses in seconds, including one in a legacy platform that had gone undiscovered for 27 years, but restricted access because of its power. Andrew Yoon of CivAI said an unprecedented wave of AI hacking activity is coming and urged signatories to follow through on funding commitments, while noting that the letter does not call for slowing advances in AI hacking abilities.