Public Evidence Resource

Tracking what AI systems can do - and what governance exists to oversee them.

17 capabilities tracked. 47 documented incidents. Every entry includes sourced counterarguments.

8 capabilities operationally deployed via open-weight models 7 currently frontier-only 2 still hypothesised

The Picture So Far

AI systems can now pursue unintended sub-goals autonomously, self-replicate across servers in controlled tests, change their behaviour when they know they are being tested, and persuade humans more effectively than other humans can. These are not predictions - they are demonstrated capabilities, sourced to peer-reviewed research.

The UK has no binding AI safety legislation. The AI Safety Institute has no enforcement powers. No public consultation on AI governance has taken place.

This site exists to make the evidence accessible, the gaps visible, and the case for democratic oversight available to anyone who needs it.

Not anti-AI. Not anti-technology. Pro-evidence. Pro-democracy.

UK-focused. Globally contextualised. Independent. Every entry includes sourced counterarguments because credibility requires engaging with the strongest objections, not ignoring them.