
VicOne Reports 28 Zero-Day Findings in Robotics Software in Under Two Days
VicOne reported 28 zero-day findings in robotics and Physical AI software in under two days, including tools associated with Google, NVIDIA and Meta. The findings show how AI-assisted research can expose weaknesses across model training, simulation and robot control before patches exist. They matter because robotics teams must assess whether such flaws could compromise model integrity, validation results or safe operation.
VicOne said the findings were identified across software and tools used to train, simulate, test and operate robots. The firm classified all 28 as zero-day vulnerabilities after initial validation, while noting the count may change after vendor and vulnerability-authority review and that some findings do not yet have CVE tracking numbers. The affected projects include software developed or maintained by Google, NVIDIA and Meta, according to the report.
The report says 75% of the findings involved development, training, simulation or validation tools, meaning most weaknesses sit before a robot is deployed. VicOne said the weaknesses could affect what a robot learns, the test results engineers trust, and the instructions sent to a robot. The firm also described a demonstration in which researchers exploited a vulnerability to prevent a robot's remote-stop communication.
The pattern shows a security gap in physical-digital systems: AI can accelerate vulnerability discovery, but the same speed can shrink response time when a flaw has no patch. Robotics tools are not isolated software; a poisoned model, tampered simulation or compromised runtime file can move through the lifecycle and change physical behavior. VicOne stressed that the scenarios are potential consequences, not demonstrated outcomes for every finding, and that impact depends on access, design and safeguards.
The findings matter because robotics teams must respond to AI-assisted discovery and assess each flaw for safety and operations. Developers, simulation users and robot operators are exposed if trusted tools become a path to model poisoning, false validation or lost control. The report connects AI capability claims to a concrete engineering test: whether autonomous systems can be built, tested and controlled without weaknesses that survive into deployment.
Robot developers, simulation teams and operators face risk if trusted robotics tools let attackers alter models, falsify tests or disrupt control. The findings show AI can speed both discovery and exposure in autonomous systems.
Sources (1)
- [1]Original source(https://spectrum.ieee.org/zero-day-robotics-vulnerabilities-vicone-security)