VicOne's 28 Robotics Zero-Days Come From a Sponsored Article With No CVE IDs and No Named Projects
Security / news
VicOne's 28 Robotics Zero-Days Come From a Sponsored Article With No CVE IDs and No Named Projects
The Tokyo security firm's count ran as paid content on IEEE Spectrum and The Robot Report. An independent March paper counted 38 flaws in three named consumer robots.

VicOne, a Tokyo-based cybersecurity company, says AI-assisted research turned up 28 zero-day vulnerabilities in under two days in software that supports robot development, and it has published no CVE identifiers, no vulnerability classes and no list of affected projects to check them against. The claim appeared on Nov. 16 as a sponsored article on IEEE Spectrum, and the same text ran on Nov. 1 as sponsored content on The Robot Report. Both are labelled as VicOne's own material.
That makes this a vendor claim, not an independent finding. What follows is what the text says, what it leaves out, and how it compares with the one independent paper on AI-driven robot hacking we could fetch.
What VicOne says it found
The article says the research covered "software and tools" in the Physical AI ecosystem, and that some affected projects are developed or maintained by Google, NVIDIA and Meta, "among others." It does not name the projects or the others. A breakdown in its Figure 1 puts the findings across the robot lifecycle.
- Development50 %
- Simulation25 %
- Deployment25 %
Source: VicOne sponsored article, IEEE Spectrum, Nov. 16, 2026 (vendor-supplied)
The caption says 75 percent involved development, training, simulation or validation tools. Only the other quarter sits in software running on a deployed robot. For a plant manager asking whether a fielded machine is exposed, that split points mostly elsewhere.

What the article does not give
The text carries no CVE IDs. It says the count of 28 "may change following vendor and vulnerability-authority review" and that some findings lack CVE tracking numbers. Every finding is labelled zero-day on initial validation only.
The attack scenarios are described as potential. A malicious pretrained model could alter training data. A tampered scene file could make a robot pass tests it should fail. Compromised motion-planning files or map data could change how a robot moves. Only one item is a demonstration: a separate VicOne demo shows a flaw that stops a robot's remote-stop command from working.
The piece also promotes two VicOne products, Radeis for finding zero-days and Rthena for runtime detection, plus a white paper. The framing favours them.
What an independent paper shows
A March 2026 arXiv paper, Cybersecurity AI: Hacking Consumer Robots in the AI Era, by Víctor Mayoral-Vilches and seven co-authors, used an open-source tool called CAI against three consumer devices: a Hookii lawnmower, a Hypershell exoskeleton and a HOBOT S7 Pro window cleaner. It reports 38 vulnerabilities found automatically, 267 or more connected Hookii devices affected and more than 3,300 internal Hypershell support emails accessed.
That paper names its targets and counts. It supports the general point that AI lowers the cost of finding robot flaws. It does not validate VicOne's 28, which concern different software.
What readers should do with it
For anyone running Google, NVIDIA or Meta robotics tooling, there is nothing to patch from this article alone. Watch for CVE records and vendor advisories, and treat unvalidated counts as provisional. The article's own advice is to find where the software is used, assess what each weakness could change, and restrict access to affected services until a fix is verified.
Robot deployments are growing, as in Rockwell Automation's AGILOX deal and Genesis AI's $500 million round, and the tooling beneath them is where VicOne says the weak points are. The next milestone is whether any of the 28 appears in the CVE database with a named project. As of Nov. 20 none has been published in the sources we could reach.
Sources
More in Security
- 01Windows Server 2028 Preview Hardens Server Core by Default in Build 29706, but NTLM Is Still OpenHundreds of settings are on from first boot, Microsoft warns Server 2025 workloads may break, and the 33 percent of servers still on 2019 are untouched.
- 02Synology's Third Critical DSM Advisory in 62 Days Hits OAuth, and the Last Patch Does Not Cover ItCVE-2026-120382 scores 9.8 and needs no login. DSM 7.4 and 7.3 builds ending -2 and -6 fix it; 7.2 and 7.1 have none yet.
- 03Codex's Sandbox Uses Seatbelt, bubblewrap and MXC, and Lets Commands Read Broadly by DefaultOpenAI's Nov. 18 engineering post says writes are confined to the workspace and the network is off, while reads stay wide open.
- 04UK Tribunal Orders Home Office to Stop Hiding Apple Encryption NoticeThe Investigatory Powers Tribunal's Nov. 19 judgment confirms the order exists. What it requires of Apple stays sealed until a hearing expected in the first half of 2027.