Fewer than 30 of 123 Physical AI companies are certifying. January 2027 is 7 months away.

The Israel Innovation Authority identified 123 Physical AI companies in its April 2026 AI strategy report. I've been reviewing their public presence — websites, publications, press releases, LinkedIn profiles — looking for one thing: evidence of an active functional safety certification effort.

My assessment: fewer than 30 of 123 (≈ 1 in 4) show any visible functional-safety certification effort. Not a completed certification. An effort. A mention of IEC 61508 or ISO 13849 on a product page. A job posting for a functional safety engineer. A press release about a TÜV assessment. Anything.

The number is low for three reasons that every early-stage team will recognize. First, cost. A full IEC 61508 certification program runs between $200,000 and $500,000. Most early-stage companies don't budget for it. Second, time. The IEC 61508 safety lifecycle adds 12 to 18 months to product development. When you're racing to ship, that's an eternity. Third, expertise. Functional safety engineers are scarce. The Deloitte 2026 State of AI report identified the skills gap as the number one barrier to AI deployment across industries. In functional safety specifically, the pool of qualified engineers is tiny relative to the number of companies that will need them.

But here's what makes this unsustainable: the EU Machinery Regulation 2023/1230 applies from 20 January 2027. That's seven months from today. There is no transition period. Any company selling AI-enabled machinery into Europe after that date must demonstrate compliance with the Essential Health and Safety Requirements in Annex III — including the new clauses on cybersecurity, control system resilience, and autonomous boundaries. For high-risk machine categories listed in Annex I Part A, a notified body must be involved.

The companies that are certifying today are, overwhelmingly, the established industrial players. Companies with 20 or more years of machinery experience, companies that went through IEC 61508 certification cycles before AI was part of the picture. They have the institutional knowledge, the relationships with certification bodies, and the internal safety teams.

The AI-native startups — the ones building the most innovative and potentially transformative Physical AI systems — are, for the most part, not certifying. They're shipping faster, iterating faster, and raising more capital. But they're building without the safety infrastructure that the regulation will require.

This creates a gap that someone needs to fill. The established certification bodies — TÜV, UL, Pilz — are building new services. UL published UL 3115 for AI safety certification in October 2025. NVIDIA launched Halos Certification for Physical AI safety. But the question remains: who builds the certification infrastructure layer specifically for the rest — the Physical AI companies that haven't started yet and don't have the in-house capability to certify on their own? That question doesn't have an answer yet.