VicOne says AI-assisted research found 28 zero-days in robotics tools

VicOne says AI-assisted research found 28 zero-days in robotics tools

VicOne, a Tokyo-headquartered Physical AI cybersecurity company, says it identified an initial set of 28 zero-day findings in less than two days using AI-assisted research. The findings are in software and tools that support the Physical AI (robotics) ecosystem, including projects developed or maintained by Google, NVIDIA and Meta, among others. The text appears on IEEE Spectrum under the line "This article is brought to you by VicOne", so it is sponsored content from the vendor, not independent reporting.

VicOne says that, based on initial validation, all 28 findings are classified as zero-day vulnerabilities. It adds that the count may change after vendor and vulnerability-authority review, and that some findings do not yet have CVE tracking numbers. Details are to become available through coordinated disclosure.

Many of the affected tools help teams train, simulate, test and operate robots. The article argues that weaknesses in them could influence the models teams build, the test results they trust and the instructions sent to robots. It says AI speeds up vulnerability discovery, which involves examining code, tracing how systems communicate and testing whether a suspected weakness can be exploited. Attackers could use similar capabilities. For a zero-day, which is previously unknown and has no available fix, faster discovery can leave teams little time to respond. That does not necessarily mean an attack has occurred, but teams still need to work out whether they use the affected software and what an attacker could access or change.

The findings span three stages of the robotics lifecycle: development and training, simulation and validation, and deployment and runtime. Per the Figure 1 caption, 75% of the findings in this research set involved development, training, simulation or validation tools, which points to risks that can arise before a robot is deployed.

The article gives one scenario for each stage. In development and training, a team downloads a pretrained model; loading a malicious file could give an attacker access to the training system and let them change training data or the model. A manipulated model could behave normally in routine testing but respond wrongly to a particular object, scene or command after deployment. In simulation and validation, importing a compromised scene or robot model could give an attacker access to the simulation environment, where they could alter simulated sensor readings, physical conditions or test results, so a robot appears to pass tests that should have exposed unsafe behavior. In deployment and runtime, an attacker could change files that influence motion planning or send malicious map data that crashes the operator's monitoring display. Compromising a remote-operation computer could also put the instructions sent to the robot at risk. The article states plainly that these are potential consequences, not demonstrated outcomes for every finding, and that actual effects depend on the attacker's access, the system's design and whether other safeguards stop changes from reaching physical behavior.

In a separate demonstration, VicOne researchers exploited a vulnerability to prevent a robot's remote-stop command from working. The article uses it to argue that teams need to test whether safeguards stay effective when the software behind them is compromised.

The piece also says a published patch still has to be proven safe. Even after a maintainer releases a fix, teams must identify affected systems, check compatibility and verify that the update preserves intended behavior. An earlier successful test remains useful evidence, but a newly found weakness may reveal conditions that test never examined. It offers three questions: where do we use the affected software (training computers, simulation environments and deployed robots), what could the weakness change (models, test results, operator visibility or robot behavior), and how do we verify protection. When immediate updating is impractical, teams may need temporary safeguards, such as restricting access to an affected service, while a validated fix is prepared.

The article closes by promoting VicOne products. VicOne Radeis is described as helping teams identify potential zero-day vulnerabilities, assess AI model and system security, and use simulation to evaluate how risks could affect robot behavior before deployment. VicOne Rthena supports threat detection and responses within approved policies during operation. Readers are also pointed to a VicOne white paper, "Why Robot Safety Depends on Cybersecurity: Protecting the Information That Safety Decisions Rely On."

Key facts

  • VicOne says it identified an initial set of 28 zero-day findings in less than two days using AI-assisted research in software supporting Physical AI and robotics.
  • Affected projects include ones developed or maintained by Google, NVIDIA and Meta, among others; the count may change after vendor and vulnerability-authority review, and some findings have no CVE numbers yet.
  • The findings span three lifecycle stages (development and training, simulation and validation, deployment and runtime), and 75% involved development, training, simulation or validation tools.
  • The attack scenarios are potential consequences, not demonstrated outcomes for every finding; a separate VicOne demo exploited a flaw to stop a robot's remote-stop command from working.
  • The text is sponsored content on IEEE Spectrum and promotes VicOne's Radeis and Rthena products and a white paper.

Why it matters

Robots are trained, tested and operated with software tools, and the article's point is that flaws in those tools can reach the physical machine indirectly: through a tampered model, a falsified simulation result or a lost view of the robot. It also pairs AI-assisted discovery with a warning that attackers could use similar capabilities. The headline figure is small in absolute terms, but it is offered as a sign of how fast AI can surface weaknesses in robotics tooling.

Who it affects

Robotics teams and manufacturers that train, simulate, test and operate robots with the affected tools, including those using projects developed or maintained by Google, NVIDIA and Meta, among others. The article says such teams need to check whether they use the affected software on training computers, in simulation environments and on deployed robots.

How to use it

The article suggests three questions: where the affected software is used, what the weakness could change (models, test results, operator visibility or robot behavior), and how to verify protection. Teams should confirm a fix addresses the weakness and that the updated system still works as intended. If an update cannot be applied right away, temporary safeguards such as restricting access to an affected service may be needed while a validated fix is prepared.

How solid is it

Weak to moderate. This is sponsored content: IEEE Spectrum labels it as brought to you by VicOne, which sells products named in the text. The 28 count rests on VicOne's initial validation and may change after vendor and vulnerability-authority review. The article does not name any specific affected project, product, CVE number or vendor-specific vulnerability. It does not say how many of the 28 findings concern each vendor. It does not name the AI tool or model used for the research, nor how the 'less than two days' was measured. It does not say whether vendors have acknowledged the findings or released patches. The article carries no independent expert commentary. The title's question, whether robotics security is keeping pace, is not answered with any measurement or comparison.

Risks and caveats

The article itself says the described scenarios are potential consequences, and actual effects depend on attacker access, system design and other safeguards. It also says faster discovery does not necessarily mean an attack has occurred, and it does not report any vulnerability exploited in the wild. The count and CVE status are provisional pending coordinated disclosure. The piece also promotes VicOne's Radeis and Rthena products and a white paper, so its framing should be read as the vendor's.

“These are potential consequences and not demonstrated outcomes for every finding.”

— VicOne, sponsored article on IEEE Spectrum