Skip to main content
ModelsResearchSecurity

DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"

Source: Ars Technica (opens in a new tab) · Ryan Whitwam

Intel Summary

Ars Technica reports that DeepMind has revealed Gemini Robotics 2.5 alongside an open safety benchmark called "ASIMOV-jailbreak." According to the report, the new robotics model achieves up to twice the success rate on certain tasks.

Why It Matters

The development links embodied robotics capability improvements directly with adversarial safety evaluation. Releasing an open safety benchmark provides researchers and developers with a public mechanism to test and measure jailbreak vulnerabilities in physical AI systems.