Standards

NVIDIA built a safety island. Is it independent — or fate-sharing?

2 min readMati Melchior
NVIDIA built a safety island. Is it independent — or fate-sharing?

Yesterday NVIDIA shipped a hardware safety layer for robots. It deserves credit: the IGX Thor module includes a functional-safety island — an independent safety processor with its own cores, power and clock, isolated from the main AI compute and engineered toward ASIL D / SIL 2. That is a real improvement over software-only safety, where the monitor and the monitored share one stack and fail together.

But functional-safety engineers have asked the same question for fifty years, and it applies here: independence is not the same as isolation. A safety monitor that shares a chip, a vendor and a design team with the system it watches shares fate with that system. Aviation, nuclear and rail each concluded that the highest integrity levels demand diversity — different teams, different substrates, different failure modes — not simply a fence drawn on the same silicon.

IEC 61508 even has a number for it: β, the common-cause factor (Annex D). β is the fraction of failures that defeat a "redundant" architecture by taking out the channels together. A low β is engineered independence; a high β is a redundancy claim with nothing behind it. Knight and Leveson showed back in 1986 that even deliberately diverse software fails to achieve true independence — and a shared physical substrate is a stronger coupling than shared design.

So the right question for any physical-AI safety architecture — NVIDIA's included — is not "is there a safety layer?" Soon everyone will have one. The question is: what is your β? What is genuinely separated — power, clock, silicon, supplier, design team — and what is quietly shared?

This is also where the open lane sits. A safety layer welded to a single vendor's compute is a powerful start, not the finish. The highest-assurance robots — and the bodies that will certify them under the EU Machinery Regulation in January 2027 — will want an independent, vendor-neutral channel that does not share fate with the AI it is judging. That layer does not have an owner yet.

Ask for the β. It is the one number that separates engineered safety from branded safety.

Share

Physical AI Safety Dispatch

Monthly analysis. No spam. One exclusive insight per issue.

One issue per month. Unsubscribe in one click from any email. Privacy policy.

Regulation

What five weeks of counting found

104 days separate 8 October 2026 from 20 January 2027, when Regulation (EU) 2023/1230 will apply as a cut-off with no transition period. Thirty-nine notified bodies held a current notification on 13 September 2026, against one hundred and forty-four under the Directive it replaces. No harmonised standard has yet been cited under the Regulation, the application guide is expected in the fourth quarter of 2026, no readiness figure has been published, and none of the seven certification bodies checked publishes how long the work takes. The deadline is measured to the day. Almost nothing behind it is.

Read →

Regulation

You have 127 days. Nobody publishes how long this takes.

127 days separate 15 September 2026 from 20 January 2027, when Regulation (EU) 2023/1230 will apply. Seven certification bodies were checked for a published duration benchmark and not one publishes it, so what follows is what is documented instead: the first assessor notified on 8 September 2024, roughly twenty-eight months before the Regulation applies, no harmonised standard yet cited under the Regulation, and seven trade associations on record that machinery was the only sector left out of the Omnibus IV transition period.

Read →

We use cookies

This site uses essential cookies to function and, with your consent, analytics cookies (Google Analytics) to understand how the site is used. Learn more.