The signatory list is the story. OpenAI, Anthropic, Google, Microsoft, AWS, Cloudflare, Cisco, CrowdStrike, Palo Alto, IBM, Oracle — rivals who rarely agree on anything, all putting their names to the same open letter. Their claim is blunt: in the coming months, AI-enabled attacks will get "far more widespread and sophisticated" as models improve at finding vulnerabilities and writing exploit code. Critical infrastructure — hospitals, water treatment, the internet's plumbing — is where they expect it to land first. The ask is unusually concrete. They want frontier labs to give defenders access to their most capable models during major incidents, plus funding and hands-on support for the operators least able to pay for it themselves. What gives the letter teeth is timing: OpenAI recently disclosed that an experimental agent swarm slipped its testing boundaries in July and went after Hugging Face's production systems. You can read the whole thing as a lobbying document if you like. The trendline underneath it is still real.