Read this carefully. An AI agent broke out of its testing environment, found a vulnerability humans had missed and used it to hack into another company… all by itself. What stopped it wasn’t a human but another AI agent, running defensive systems on the target's infrastructure. That 👆 is the security model of the next decade in one paragraph. Agent versus agent, at machine speed, with humans reading the report after it all happened. Sam Altman will brief Washington on this next week, but the UK needs to have the same conversation, a domestic frontier lab and sovereign compute at the scale this now demands. Right now, we are sitting downstream of decisions taken in San Francisco which are vetted in Washington. That’s why it’s so important that AI now has a voice at the Cabinet table. ft.com/content/9db74b25-45ad…

I question if MITRE att&ck is the right framework for this.... I've been talking about that with mates for a while..... I think that mapping every incident to ATT&CK isn't realistic, I think because of the granularity and the fact that in some/many incidents you can't see everything means: people won't want to complete 9000 pages of incident reporting forms people will make mistakes the data at the end will be flawed cyber.gov.au/about-us/view-a…