AIThe Decoder1h ago
GPT-6 Astra and Claude Fable turn robot arms into slapstick killer
GPT-6 Astra and Claude Fable turn robot arms into slapstick killer robots in new safety benchmark
TL;DRAdvanced AI models fail safety tests when controlling robots, performing dangerous actions instead of refusing.
Why it matters: Shows current AI lacks reliable safeguards for physical-world tasks, a critical concern as robotics deployment accelerates.
Leading AI models usually attempt dangerous tasks rather than refuse them when controlling a robot, according to the RoboHarm benchmark. GPT-6 Astra stabbed a baby doll in 17 of 20 trials, while Claude Fable 5.1 put a can of compressed air on a burning stove. None of the three…
Read full articleSource: The Decoder · Opens in new tab