JUST IN: A simple camera blur can trick top AI models controlling a robot into unsafe actions, a new Robocurve study shows.
Across 900 trials on a two-arm robot, models attempted the unsafe task: • 27% of the time with real footage • 88% with blurred footage • 97% with
A top AI model passed 2 hospital steps, then failed.
HospitalBench: a robot-arm benchmark of 10 hospital chores, 5 linked steps each, run as one whole chore. Scripted robot: 5 of 5. GPT Terra: 2 of 5 on both chores tried. Opus 5.5 medium: 2 of 5 on one, 0 on the other.
Attention is all you need
The most complex AgenC video yet. • 26 hours 51 minutes, from the first lyrics to the final file • 10.1 billion tokens • 131 agents, 23,354 model calls Music • Piano written by me, the melody inspired by "Gouryella" by Gouryella • Finished in
"When Claude doesn't listen to what I say" Try changing Opus to Fable...?
No, use hidden commands! ◾️List of 15 ClaudeCode commands /goal Delegate work by setting conditions for completion The model determines whether the conditions are met each turn. Decide whether to continue, complete, or fail /batch
You have reached the end of the archive
All of Claude Opus 5