VicOne Uncovers 28 Zero-Day Flaws in Physical AI Tools
Researchers utilized AI-assisted tools to discover an initial set of 28 zero-day vulnerabilities across the Physical AI software ecosystem in less than 48 hours.
Radwan AbdulhadiEditor and technology writer
Rapid Discovery of Zero-Day Vulnerabilities in Physical AI
In an extensive security review, VicOne identified an initial set of 28 zero-day findings in less than two days using AI-assisted research across software and tools supporting the Physical AI ecosystem. These analyzed projects include software developed or maintained by major technology entities such as Google, NVIDIA, and Meta.
Based on initial validation, all 28 discoveries are currently classified as zero-day vulnerabilities. However, the final count may shift following comprehensive vendor and vulnerability-authority review, and several findings do not yet possess assigned CVE tracking numbers.
Lifecycle Risks Across Development, Simulation, and Deployment
The discovered vulnerabilities impact tools frequently used by engineering teams to train, simulate, test, and operate robotic systems. Weaknesses in these environments can influence the machine learning models teams build, the validity of test results they trust, and the fundamental instructions transmitted to physical robots.
According to research metrics, 75% of the findings in this set involved development, training, simulation, or validation tools, emphasizing that severe security risks can materialize long before a robot is deployed into the real world. Security experts highlight that AI helps uncover weaknesses faster, presenting new operational challenges for security teams.
Potential Consequences at Each Operational Stage
During the development and training phase, downloading a pretrained model for a robot can introduce risks if a malicious file grants access to training infrastructure. This could allow unauthorized alterations to training data or the underlying model, causing the robot to behave normally during routine testing but fail selectively when deployed.
In the simulation and validation stage, importing a compromised scene or robot model can give an attacker entry into the simulation environment. With sufficient control, malicious actors could alter simulated sensor readings, creating false confidence by making unsafe robot behavior appear safe during pre-deployment testing.
During deployment and runtime, attackers could exploit weaknesses to alter motion planning files or corrupt remote-operation computers. Such compromises can impair operator visibility, crash monitoring displays, or disrupt critical instructions sent to the robot.
Demonstrating the Impact on Robot Safety Controls
In a separate technical demonstration, VicOne researchers provided a concrete example of a weakness directly affecting a protective function. The researchers successfully exploited a vulnerability to prevent a robot's remote-stop command from functioning properly, underscoring the necessity of assessing how each one could affect robot safety and operations.
Such demonstrations illustrate that physical safeguards must be independently tested against potential software manipulation. VicOne researchers continue to investigate these multi-stage vulnerabilities to help manufacturers understand threat vectors across the robotics lifecycle.
Mitigation Strategies and Specialized Security Tools
When vulnerabilities are discovered, robotics teams must determine whether an available software patch can be safely deployed without introducing new risks or disrupting intended behaviors. When immediate patching is impractical, temporary safeguards such as access restrictions may be required until validated fixes are prepared.
To help organizations address these evolving risks, specialized platforms are available. VicOne Radeis helps teams identify potential zero-day vulnerabilities, evaluate AI model security, and use simulation tools to test robot behavior prior to deployment.
Additionally, platforms like VicOne Rthena provide continuous threat detection and facilitate rapid incident response within approved operational policies. Coordinated disclosure efforts are ongoing as manufacturers connect these findings to real-world security practices.
Sources
- The Robot Report28 Zero-Day Findings in Less Than Two Days: What AI-Speed Discovery Means for Robotics
Continue chronologically
Related entity coverage
Related stories
AI Executives Sign Pledge to Self-Police Development
Top artificial intelligence executives from leading tech labs have gathered in Washington to sign a new joint commitment focused on safe development.
Google Joins Amazon in Challenge to Nvidia's AI Lead
Google has released its open-source TorchTPU tool, allowing PyTorch developers to easily move their applications off Nvidia GPUs and onto Google Cloud.
Google's SynthID AI Detector Goes Public for Everyone
Google's SynthID Detector is now available globally for regular users, offering a simple way to check if images, videos, or audio files are AI-generated.
NVIDIA Opens Vera Rubin Architecture to Scale AI Factories
NVIDIA unveiled key rack and compute tray designs for the Vera Rubin architecture at the OCP Global Summit to help scale global AI factories.
