Editorial standards

Notes & Corrections

Factual errors are corrected within a fixed window. Every correction is logged here, dated, and attributed.

Corrections policy

Factual errors in any post, dispatch, paper, or book published on this site are corrected within 7 days of receiving evidence. The correction is published below, dated, and links back to the original surface.

Anyone — readers, subjects of coverage, regulators, journalists — is invited to submit evidence. The correction will identify the change made and (where appropriate) cite the source of the evidence.

Policy effective from 2026-05-09.

Contact: mati@physical-ai-safety.com

Correction log

Post

Zero robot insurance products claim corrected: China shipped dedicated policies in 2025

The post claimed no major global insurer had launched a standalone robot safety insurance product. In late 2025 Chinese insurers shipped dedicated standalone policies (China Pacific Property; CPIC Ji Zhi Bao; PICC P&C). Title, excerpt and the Zero section were rewritten: first movers are Chinese while no US/EU insurer has yet shipped one, and the safety-data gap still constrains pricing everywhere.

Evidence
Humanoids Daily (2025); China Daily, 12 Dec 2025; TMTPost; PICC P&C and China Pacific announcements.
Post

The Gap in Physical AI (April 2026) annotated with a June 2026 update on NVIDIA Halos

A dated Update — June 2026 section was appended (April text left intact) noting that NVIDIA's 22 June 2026 Halos for Robotics launch validates the core argument and moves the gap to the independent, vendor-neutral safety layer, linking the updated safety-infrastructure-landscape post.

Evidence
NVIDIA Newsroom, 22 June 2026; cross-link to /blog/safety-infrastructure-landscape-who-builds-what.
Post

Safety-infrastructure landscape: the empty lane filled by NVIDIA Halos for Robotics

The post described full-stack safety for AI-driven robots as a product category that did not yet exist. On 22 June 2026 NVIDIA announced Halos for Robotics (safety-capable compute with an independent on-chip safety processor, safety software, accredited inspection lab; 40+ partners; Agility Digit first). The post was updated: the lane now has its first major occupant, and the remaining gap is reframed as a vendor-neutral, independent safety layer. The thesis — high-integrity safety needs independence from the system judged — is unchanged.

Evidence
NVIDIA Newsroom, 22 June 2026; NVIDIA IGX Thor functional-safety island; NVIDIA developer technical blog.

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.