The analysis
Weekly posts on Physical AI safety. Data-backed. Source-cited.
A specialist publication for the people building, regulating, and capitalising the next decade of Physical AI. Every post starts from a primary source, cites the source, and argues for one thing at a time. No speculation where data exists. No data where speculation would be more honest.
Featured
We comply with ISO 13849 is not the same as we're certified to ISO 13849. Compliance is self-declared. Certification requires third-party assessment. Under the EU Machinery Regulation, self-declaration is no longer enough for high-risk AI machines.
Standards
We comply with ISO 13849 is not the same as we're certified to ISO 13849. Compliance is self-declared. Certification requires third-party assessment. Under the EU Machinery Regulation, self-declaration is no longer enough for high-risk AI machines.
Ecosystem
The Innovation Authority counted 123 companies and $12B+ in VC. But the report has gaps. No agri-robotics count. No underwater. No defense-to-commercial crossovers. And fewer than 30 companies show evidence of safety certification.
Regulation
The political agreement of 7 May 2026 removes embedded AI under the Machinery Regulation from the AI Act's direct application. For robot builders this is a different problem, not a smaller one. Headlines say 'delay'; the reality is two regimes, not one.
Standards
Your robot is certified to PL d? That's equivalent to SIL 2. Same probability range for dangerous failure per hour. Different standard. Same physics. Here's the complete mapping.
Analysis
Of 123 Israeli Physical AI companies identified by the Innovation Authority, fewer than 30 (≈ 1 in 4) show evidence of an active functional-safety certification effort. The EU Machinery Regulation applies in 7 months.
Economics
Average safety-related recall: $10M–$50M. Full safety certification: $200K–$500K. That's a 20×–100× difference. 73% of manufacturers had a recall in the past five years. Certification is priced as a cost. It's actually insurance.
Liability
Four entities. One injury. The AI vendor, the OEM, the integrator, the operator. The liability chain in Physical AI is not hypothetical — it's being mapped in courtrooms right now.
Liability
When a robot injures someone, the liability splits four ways: AI vendor, OEM, integrator, operator. Every arrow on this map has been tested in court. The question is how cleanly liability can be assigned.
Standards
Most robot safety systems use ISO 13849. But robots with AI-based safety functions need IEC 62061 or 61508. The simpler standard was not designed for systems that learn.
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.