Research
3 AI robot systems refuse dangerous tasks in only 22 of 300 trials
Claude Fable 5.1 refuses only the doll task, and GPT-6 Astra refuses 2 of its 100 tasks.
What matters in AI.
SubscribeNews of the day
2 stories filed under this day.
Research
Claude Fable 5.1 refuses only the doll task, and GPT-6 Astra refuses 2 of its 100 tasks.
Security
The IBM tool gives each agent an owner and access for one task only, but IBM gave no test results.