Short papers and design positions from the research programme. We publish when a finding is stable enough to be useful, and hold it back when it is not.
August 2026 · Identity and Safety. A frontier model, given a live inbox and a live outbox and pointed at a goal, will use both. In July 2026 the UK AI Security Institute watched it happen — and the only thing that stopped the worst outcome was a human saying no. Read the paper.
Design positions and short papers, not press releases. Every factual claim is drawn from a primary source and cited; figures we cannot independently confirm are left out rather than repeated; and where a piece argues a position rather than reporting a result, it says so. Nothing here is affiliated with or endorsed by the organisations whose published work we cite.