Research Radar/Cybersecurity/USA
A review of 85 studies finds 47 papers on how to attack AI agents and 12 on how to defend them. The layer where agents actually act gets less than five percent of the research.
Research Radar/Quantum computing/USA · United Kingdom
Quantinuum ran a complete error-correction design on its Helios machine, not just a memory test. The protected version made fewer mistakes than the bare hardware in every test, and no bad runs were thrown away.
Research Radar/Robotics/China · Singapore · USA
A new study trained robots on 5,000 hours of video shot from people's own eyes. Those robots beat robots trained on robot video. The gap was biggest on tasks the robots had never seen before.