AIM Intelligence launches Physical AI Safety Consortium

AIM Intelligence has launched PASC, a global consortium focused on safety standards for robots and other physical AI systems. The group includes about 30 researchers and industry practitioners from Google, Microsoft, Oracle, LG Electronics and universities worldwide as it begins work on definitions, benchmarks and evaluation methods.

Why it matters: - Physical AI systems act in the real world, so failures can cause direct harm, not just bad text outputs. - PASC is trying to define common safety standards before the field scales further. - The consortium could help shape future evaluation methods and standardization for robots and autonomous systems.

What happened: - AIM Intelligence launched PASC, short for Physical AI Safety Consortium, on Sept. 17, 2026. - The consortium is global, open and convened by AIM Intelligence on a not-for-profit basis. - The launch brings together experts from Google, Microsoft, Oracle and LG Electronics, along with researchers from universities and AI safety institutes worldwide. - AIM Intelligence CEO Sangyoon Yu announced the launch. - PASC includes about 30 researchers and industry practitioners.

The details: - PASC covers physical AI systems such as robots, autonomous systems and industrial equipment. - The consortium is focused on identifying, assessing and controlling risks before physical actions cause harm. - PASC aims to establish common definitions of physical AI safety, evaluation methodologies, benchmarks and research foundations. - Individuals and institutions can contribute research, models, hardware, compute resources and data. - Key participants include Jenny Ni of Google; Adrian de Wynter of Microsoft and the University of York; Hitesh Laxmichand Patel and Amit Agarwal of Oracle; Jiyong Jang of LG Electronics; Myuhng-Joo Kim of the Korea AI Safety Research Institute; Sicun Gao of UC San Diego; Sharon Li of the University of Wisconsin; and Yaqi Xie of the University of Illinois. - PASC’s first joint project is a position paper on the definition and evaluation direction of physical AI safety. - The consortium also plans joint evaluation systems, benchmarks, adversarial testing in operating environments and validation research in simulation and real robot environments. - The group says safety must be continuously observed, updated and intervened on rather than treated as a one-time determination. - PASC’s research questions include whether risks can be defined well enough, whether warning signs can be detected early enough to intervene, whether results stay valid across changing environments and robot morphologies, and whether safety systems can withstand adaptive attacks. - Haon Park, CTO of AIM Intelligence, said PASC is meant to create a shared language and a foundation for validating physical AI safety. - More information is available at PASC participation and research.

Between the lines: - The consortium is positioning physical AI safety as a field that needs its own standards, separate from traditional language-model safety. - The emphasis on open contributions and not-for-profit operation suggests AIM Intelligence wants PASC to function as a shared research platform, not a proprietary program. - The focus on real-world testing, simulation and robot morphologies signals that safety in this space will depend on both technical benchmarks and changing deployment conditions.

What's next: - PASC will publish a position paper on physical AI safety definitions and evaluation. - The consortium plans to broaden into benchmark design, adversarial testing and validation research. - PASC will likely seek more participants and contributions as the research agenda expands. - The group may also influence future standardization efforts if its frameworks gain traction.