> ## Content Index
> Fetch the complete content index at: https://www.humanoidthreats.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Humanoid Threats Brief #5: AI Bug Hunting Reaches Robot Toolchains
- URL: https://www.humanoidthreats.com/humanoid-threats-brief-5/
- Published: 2026-11-19T06:58:58.000Z
- Updated: 2026-11-19T07:00:00.000Z
- Description: This week's humanoid robot security brief: a claim of 28 AI-found robotics zero-days, China's new training-data standards and EU liability rules ahead.
- Author: Mira Castell
- Tags: The Brief, News

This week: a security vendor says AI helped it find 28 zero-days in robotics software in under two days, China publishes new standards for humanoid robot training data, and the EU's product liability rules for software flaws are now three weeks away. No new robot CVEs or confirmed incidents were published this week, so those sections are skipped.

## Top threat: AI-speed bug hunting reaches robot toolchains

Security firm VicOne says it found "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", including projects "developed or maintained by Google, NVIDIA, and Meta". It says 75% of the findings sit in development, training, simulation or validation tools rather than on deployed robots, and that its researchers "exploited a vulnerability to prevent a robot's remote-stop command from working" (sponsored article in [IEEE Spectrum](https://spectrum.ieee.org/zero-day-robotics-vulnerabilities-vicone-security?ref=humanoidthreats.com), November 16).

Treat the numbers with care. Both write-ups so far are sponsored content, the affected projects and the robot in the demo are not named, no CVEs have been published, and VicOne itself says "the count may change following vendor and vulnerability-authority review".

Why it matters (analysis): if the claim holds up, the risk is less about one robot and more about the pipeline. A flaw in a simulator or training tool can quietly change the models and test results that safety sign-offs rely on. And a remote stop that can be blocked is exactly why a humanoid needs a hardwired stop independent of its software. Our new explainer: [What is a robot emergency stop?](https://www.humanoidthreats.com/what-is-robot-emergency-stop/)

## Policy watch

- **China sets rules for humanoid training data:** China's Ministry of Industry and Information Technology approved 274 industry standards in Announcement No. 30, dated November 10 and published November 16\. They include six humanoid and embodied-AI standards, one covering the full lifecycle of humanoid robot training data from collection to destruction, plus a smart-home standard on protecting audio and video data. The humanoid standards take effect on publication. [MediaNama](https://world.medianama.com/china-approves-274-industry-standards-covering-humanoid-robots-ai-semiconductors-and-data-security/?ref=humanoidthreats.com). Background: [Humanoid robot privacy](https://www.humanoidthreats.com/humanoid-robot-privacy/).
- **EU software liability starts December 9:** the new Product Liability Directive applies to products placed on the EU market after 9 December 2026 and says a cybersecurity vulnerability can make a product defective. [EUR-Lex](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L%5F202402853&ref=humanoidthreats.com). Our explainer: [Who is liable when a hacked robot causes harm?](https://www.humanoidthreats.com/who-is-liable-when-a-hacked-robot-causes-harm/)
- **FCC Covered List exemptions:** an FCC Public Safety and Homeland Security Bureau notice recorded on November 6 announced the "conditional approval and exemption of certain uncrewed aircraft systems, routers, advanced robotic devices, and power inverters" from the Covered List ([Digital Policy Alert](https://digitalpolicyalert.org/event/45565-federal-communications-commissions-public-safety-and-homeland-security-bureau-updated-the-covered-list-to-reflect-conditional-approval-of-certain-advanced-robotic-devices-and-power-inverters?ref=humanoidthreats.com)). The specific robots are not in the public summary yet. Background: [FCC robot ban explained](https://www.humanoidthreats.com/fcc-robot-ban-covered-list-humanoid-robots/).

## Tool of the week: test the stop path, not just the button

A practice rather than a product. Once a quarter, trigger each robot's emergency stop from every channel it has (the button on the robot, the wireless pendant, the app) with the robot's network cut and its main computer busy. A stop that only works when the software is healthy is not an emergency stop. Write down which IEC 60204-1 stop category it performs and whether the humanoid stays upright. Details in our [E-stop explainer](https://www.humanoidthreats.com/what-is-robot-emergency-stop/) and the [security checklist](https://www.humanoidthreats.com/humanoid-robot-security-checklist/).

## Directory pick: Alias Robotics

If AI-assisted bug hunting is the story of the week, Alias Robotics is one of the teams that applies it to robots. The Basque company started in robot security and now builds CAI, an open-source framework for AI security agents, alongside its Robot Immune System endpoint protection. Read our [Alias Robotics profile](https://www.humanoidthreats.com/directory/alias-robotics/) or browse the full [security directory](https://www.humanoidthreats.com/directory/).

---

Know someone who should read this? Forward it, or send them [humanoidthreats.com/newsletter/](https://www.humanoidthreats.com/newsletter/)

Get the free [Humanoid Robot Security Checklist](https://www.humanoidthreats.com/humanoid-robot-security-checklist/).

Get the weekly Humanoid Threats Brief → [Subscribe](#/portal/signup)