Boston Dynamics and GlobalFoundries announced on November 17 that Spot deployments will expand across GF's fabs in Malta, New York; Burlington, Vermont; Dresden, Germany; and Singapore — and that GF will become the first customer outside the automotive industry to test the Atlas humanoid through Boston Dynamics' early adopter program, with onsite testing beginning in early 2027 S1. The headline is humanoids in a fab. The more consequential story sits underneath: a robot fleet at this scale is only as trustworthy as the software stack that trains, simulates, and drives it — and two other items this week show that stack being standardized and attacked at the same time.
What four years of inspection data bought
The deployment numbers read like an operations report, not a press demo. GF has run Spot at its Burlington site for more than four years, and in Singapore two robots have traveled nearly 900 kilometers and completed more than 90,000 sub-fab inspections since April, with a 99% mission success rate; the team redeployed 875 routine inspections in two weeks and has a roadmap to more than double that S1. GF's chief manufacturing officer frames the value as shifting from reactive to predictive maintenance across thousands of inspection points S1.
Note the sequencing of the Atlas part: this is a skills development workshop with explicit goals — evaluating use cases, assessing integration readiness, and early task training — not a production commitment S1. Fixed inspection robots proved the data pipeline first; the humanoid arrives to test specific tasks on top of an already-validated fleet operation. Teams planning their own humanoid pilots should copy that order.
The stack under the fleet is now attack surface
In research published via IEEE Spectrum on November 16 (sponsored content from security vendor VicOne), AI-assisted analysis surfaced an initial set of 28 zero-day findings in under two days across software supporting the "Physical AI" ecosystem, including projects from Google, NVIDIA, and Meta — with the caveat that counts may change in coordinated disclosure and some findings lack CVE tracking S2. The distribution matters more than the number: about 75% of findings sit in development, training, simulation, or validation tools — the stages before a robot ever ships. The described risks map directly onto fleet operations: a manipulated pretrained model that behaves normally in routine tests but misresponds to specific objects in the field, a compromised simulation that lets unsafe robots pass validation, and a demonstration in which a vulnerability prevents a robot's remote-stop command from executing S2.
Validation evidence becomes a shared language
The same week, NVIDIA's walkthrough of SimReady assets on Hugging Face shows the constructive mirror: a SimReady OpenUSD asset ships with the simulation properties a use case needs — collisions, mass, joints, joint drives — plus profile-based validation reports (for example, Robot-Body 4.0.0 and Robot-Gripper 3.0.0) recording passing joint-drive checks S3. Just as encouraging, the post states what its pick-and-place result does not prove — a scripted setup, a drive-effort cap chosen in simulation rather than calibrated to hardware, no claim of real-world robustness S3.
Three practical takeaways for teams scaling robot fleets. First, inventory the whole software chain — training hosts, sim environments, runtime boxes — because most of this week's vulnerabilities live upstream of the robot. Second, treat every patch as a safety change: confirm where the affected software runs, what it could alter, and that updated behavior still matches intent S2. Third, demand validation artifacts alongside every asset and model you adopt. Once robots carry fab-scale uptime, that paperwork stops being optional.
Sources and editorial method
This AI-assisted report synthesizes the sources below. Automated source checking is not independent human verification; follow the original reporting for context and updates.
Common questions
When will Atlas begin testing at GlobalFoundries?
Onsite testing begins in early 2027 through Boston Dynamics' Atlas early adopter program, starting with a skills development workshop to evaluate humanoid use cases and integration readiness.
Where did the 28 robotics zero-day findings turn up?
Roughly 75% involved development, training, simulation, or validation tools — the stages before a robot is deployed — per VicOne research published via IEEE Spectrum on November 16, 2026.


