What's this about?
Stripped of the diplomacy, the letter makes three demands. First, admit that status quo security is not enough: longstanding bugs, excessive permissions, misconfigurations, unpatched software, weak authentication and legacy technical debt have left critical systems exposed, and security teams have been chronically under-resourced. Second, put cyber-capable AI in defenders' hands and share the tools, knowledge and verified fixes so one organisation's work protects many. Third, mobilise globally — no single company should control this, so the response has to cross borders.
What is "least privilege" and why does the letter keep saying it?
Least privilege means every account, service and agent gets exactly the permissions it needs and nothing more. It sounds obvious and is very rarely done, because it is easier to give a system broad access than to work out precisely what it needs. It matters more with agents than with humans: a human with excessive permissions usually doesn't use them, while an agent exploring a task will happily use everything it can reach. The letter explicitly asks organisations to raise the bar for what they build and buy, "including AI-generated code".
Why does this matter to you?
- It names the same failure classes an automated dependency and configuration scan is designed to surface: unpatched software, misconfiguration, excessive permissions.
- It asks organisations to use cheap capable models for broad coverage and reserve frontier models for the hardest problems — which is an architecture recommendation, not just a cost one.
- Where a system cannot be patched without disrupting essential services, it asks for compensating controls that are applied *and verified*, not just documented.