AI-assisted audit turns up 28 zero-days across robot development tools

VicOne says AI-assisted research found 28 zero-day flaws in Physical AI software in under two days. The report is sponsored and the count may change.

Security firm VicOne reports that AI-assisted research turned up 28 zero-day findings in under two days. The affected software and tools support the Physical AI ecosystem, including projects developed or maintained by Google, NVIDIA and Meta, among others. Initial validation classed all 28 as zero-days, and some have no CVE tracking numbers yet.

The findings span three stages of a robot's life: development and training, simulation and validation, and deployment and runtime. About 75% involve development, training, simulation or validation tools, so the risk can arise before a robot is ever deployed. One scenario is a malicious pretrained model file that lets an attacker alter training data, so the robot behaves normally in tests but wrongly on a particular object, scene or command. Another is a compromised simulator asset that falsifies test results so an unsafe robot passes validation. At runtime, an attacker could change motion-planning files or crash an operator's monitoring display. In a separate demonstration, VicOne researchers exploited a flaw to stop a robot's remote-stop command from working.

This matters for abundance because cheap, widely deployed robots only help if people can trust them. Flaws found at AI speed, before deployment, make safe robotic labor in goods and care more plausible. The same speed will also be available to attackers, so defenders have to keep pace.

The caveats are significant. This is sponsored content that promotes VicOne's own products, Radeis and Rthena. The count of 28 may change after vendor and vulnerability-authority review. The consequences described are potential, not shown for every finding, and real effects depend on attacker access and system design. The article does not report any attack, and it does not say which products or vendors have which findings. VicOne suggests teams check where affected software is used, what a weakness could change, and how to verify a fix, and restrict access to affected services if updates must wait.

Source: The Robot Report