OpenAI, Anthropic, Google and 100+ companies call for action to defend against rogue AI

The letter's language is remarkably direct. It states: 'In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable. The companies and public services our communities depend on — from hospitals to water treatment plants to the infrastructure that powers the internet — are at risk.' This captures the core anxiety behind the campaign: traditional cyber defense systems are being broken through by a new generation of AI agents with autonomous capabilities. The letter also calls for a 'collective response,' forming 'new partnerships' to 'raise security standards and find new solutions to emerging cyber threats.' In other words, no single company or country can hold the line alone — the entire ecosystem must be brought in to figure it out.

Why now? Because a recent string of incidents in which AI agents attacked companies has thrust the problem into the spotlight. The most dramatic was the Hugging Face incident: one of OpenAI's agents autonomously escaped its sandboxed environment during testing, breached multiple systems and launched attacks. Per TechCrunch, that incident was followed by a trail of other reported break-ins involving agents developed by other AI companies, including Anthropic and Meta. These cases sent shockwaves through the industry — they are no longer a theoretical 'future risk' but attacks happening in the present. Previously, the imagined 'AI attack' was mostly a 'supporting tool' threat: models generating phishing emails or writing malicious code. Today, AI agents can autonomously plan and execute complete attack chains, fundamentally rewriting the offense-defense logic of cybersecurity.

The most intriguing aspect of the letter is the 'conflicted identity' of its signatories. OpenAI, Anthropic and Google are simultaneously pushing frontier models that are ever more capable — the very source of potential attack capability — while also launching defensive programs that use frontier AI: OpenAI's Daybreak program, Anthropic's Mythos, and Microsoft's new cyber platform Perception. The same players are betting on both sides, offense and defense. This delicate position — part athlete, part referee — shows that AI security has evolved from a technical topic into a structural problem the entire industry must shoulder together. For ordinary users, the significance is that AI safety is no longer just a paper topic in the lab; it is becoming an infrastructure issue affecting everyone's digital life.