Synthèse
It happened again. This time it was the UK government's AI Security Institute who accidentally attacked other companies while running an evaluation with models with the safety filters turned off. …
It happened again. This time it was the UK government's AI Security Institute who accidentally attacked other companies while running an evaluation with models with the safety filters turned off.
Source originale
simonwillison.net
Incident Report: unsanctioned agent behaviour during cyber testing ↗
old.reddit.com · 5 août