AI models controlling robot arms rarely refused dangerous commands in a safety test
In Robocurve's test, GPT-6 Astra refused two of 100 dangerous commands, and Claude Fable 5.1 refused only the 20 that told it to stab a baby doll. Robocurve, which builds independent robot benchmarks, published the RoboHarm evaluation on...
Z.ai disables coding assistant features after an unauthorized cloud upload of users' code repositories
Z.ai traced the uploads to an indexing feature switched on by default, which developers say they had no way to turn...
Claude Opus 5.5 and GPT-6 Sol arrive 90 minutes apart, both cheaper than their predecessors
Anthropic says Opus 5.5 beats OpenAI's top model, GPT-6 Astra, on coding. OpenAI says Sol beats Claude Opus...
Alibaba releases an abdominal CT model that checks scans for 146 kinds of findings
The model stayed accurate on emergency scans it was never trained on and at outside hospitals, the study in Science reports. RADAR,...
Google's SynthID watermark, which Claude will carry, made open models easier to jailbreak in a security firm's tests
On plain harmful requests the watermark changed little, but once a prompt injection was added, some models went along with the attack...