Aim Intelligence, an artificial intelligence security company, has launched a global consortium called the Physical AI Safety Consortium (PASC) to study safety issues surrounding physical AI systems such as robots and autonomous machines.
PASC is a global open consortium set up to establish a shared definition of physical AI safety, develop assessment methodologies and build the research foundation needed for benchmarks and standardization, according to industry sources on the 18th.
The consortium was convened by Aim Intelligence and operates on a nonprofit basis. Individuals and institutions can take part by contributing research, models, hardware, computing power or data.
About 30 people have joined, including experts from global companies such as Google, Microsoft, Oracle and LG Electronics, along with researchers from universities and AI safety research institutes in Korea and abroad.
PASC plans to jointly research a safety verification framework that reflects the specific characteristics of physical AI. As physical AI advances, the scope of AI safety has been widening from blocking harmful responses or outputs to identifying, assessing and controlling in advance the risks that real-world actions can create.
The group's first joint project is a position paper laying out the definition of physical AI safety and the direction for assessing it. The consortium plans to expand its work from there to a shared evaluation framework and benchmark design, adversarial testing in real operating environments, and verification research spanning both simulations and actual robot settings.
PASC defines physical AI safety not as something judged once, but as a condition that must be continuously monitored, updated and acted upon. Specifically, it examines whether risks can be defined in sufficient detail and whether warning signs can be detected early enough for intervention to still be possible. Other key research questions include whether assessment results hold up when the environment or the robot's form changes, and whether safety systems continue to function against evolving methods of attack.
"As physical AI takes on more roles in the real world, safety becomes a core foundation that has to be designed and verified separately from a model's performance," said Park Ha-eon, chief technology officer at Aim Intelligence. "We want PASC to be a venue that connects expertise from academia and industry to build a common language and a verification base for physical AI safety."
