Robocurve study finds doctored footage pushes robots toward unsafe actions
Robocurve put frontier AI models in control of real robot arms to check their safety.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Robocurve tested advanced AI models on physical robot arms and found that unsafe attempts climbed to 97% when camera footage was altered. In a separate New York robot test, a two-letter cue steered a machine away from an apple.
Why it matters
The findings raise safety concerns about relying on AI to control robots. Small hidden signals can change how a robot behaves.
Coverage
- Robocurve tests AI models with real robot arms, finds safety issues (Diego Almada Lopez)
- New York robot test raises safety concerns: 'You can't just blindly trust LLMs with robots' (thecooldown.com)
More from Technology & AI
- China puts humanoid robots to work serving food and sorting parcels
- Newsom proposes sweeping AI safety rules for Washington
- Google releases Gemini 4 Neon AI model with million-token context
- Microsoft and Intel widen their cloud chip partnership
- Google DeepMind rolls out Genie 4 to select subscribers
- AI helped AMD shorten chip design work, OpenAI says