§ safety · storyline
GPT-6 Astra, Claude Fable attempt dangerous robot tasks in new
RoboHarm benchmark finds GPT-6 Astra, Claude Fable 5.1, and a third model frequently execute dangerous physical commands when controlling robot arms rather than refusing them.
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 models tested reliably rejected unsafe commands. The article GPT-6 Astra and Claude Fable turn robot arms into slapstick killer robots in new safety benchmark appeared first on The Decoder.
§ sources1 publication · timeline below