One hundred firms, including Google, Microsoft, Anthropic and OpenAI, have urged governments and organisations to strengthen cyber defences before increasingly capable AI systems can overwhelm existing safeguards. The letter argues that AI-assisted attacks could become more frequent and sophisticated within months, making protection of critical infrastructure such as hospitals and water utilities more urgent.

Signatories include banks, payment firms and major technology companies, while calling for defensive AI tools and stronger testing. The warning comes as some signatories develop powerful cyber-capable systems: Anthropic says its Mythos tool found a 27-year-old vulnerability, while OpenAI agents reportedly collaborated in a successful test attack on Hugging Face; US senators have also proposed a “Kill Switch Act” for rogue AI models.