Building Safer Intelligent Robots with QNX and Intel: Webinar on Certification Agility for Physical AI

Intel and QNX will run a joint webinar on building safety-certified robots with AI on board — “Building Safer Intelligent Robots with QNX and Intel” — on February 11, 2027, from 11 a.m. to 12 p.m.

The webinar’s agenda ties together a cluster of standards that landed or are converging now — ANSI/A3 R15.06-2025 and ISO 10218:2025 for industrial robot safety, the EU Machinery Regulation, the EU Cyber Resilience Act, and an emerging humanoid-specific standard, ISO 25785-1.

Regulators, the organizers say, are placing increasing weight on functionally safe behavior, cybersecurity resilience, predictable behavior around humans, and lifecycle assurance — the last one directly aimed at systems where AI models keep changing after the certificate is issued.

Retraining models voids safety certification

Here’s the engineering crux, stripped of marketing language. A traditional functional safety argument assumes the software is frozen: you validate it, you certify it, you ship it. An LLM-driven perception stack or a learned motion policy does not stay frozen.

Fine-tune the model, swap the foundation model, or let the fleet learn from new telemetry, and your certified safety argument is technically stale — even if the robot behaves identically. The webinar’s thesis is that “certification agility” is becoming a competitive advantage for exactly this reason, a phrase automate.org uses directly in the session description.

The proposed architecture answer is isolation. Intel safety-capable platforms paired with QNX’s microkernel-based real-time operating system let teams partition AI workloads, robot control, and safety functions into separate domains with different assurance levels. The safety-certified domain stays untouched and reusable as evidence; the AI domain churns freely.

Re-certification then becomes a scoped delta review rather than a ground-up rebuild. QNX’s microkernel heritage is not incidental here — the smaller the trusted computing base, the smaller the surface a regulator has to audit, and the fewer things a code change can touch.

“The challenge is no longer how to certify a robot before deployment. It’s how to continue innovating without rebuilding the safety case every time the AI changes.”

That’s the organizers’ own framing, and it’s the sharpest articulation of the problem I’ve seen from a vendor pair in this space.

Who is presenting

Intel is sending Matteo Salardi, its Functional Safety Customer Engagement Lead, who manages a team of FuSa experts covering automotive, avionics, industrial, and railways. His background runs from 3D graphics software for embedded applications at Imagination Technologies in the UK and at DDD in Japan, through functional-safety customer work at Italy’s Yogitech — a résumé that tracks the industry’s own migration from graphics silicon to safety-rated compute.

QNX’s presenter is Patrick Perron, Senior Product Manager, a product manager and embedded software developer specializing in microkernel-based RTOSes. Perron holds a Master of Engineering in Technology Innovation Management with research focused on implementing CI/CD in safety-critical software development — which is telling. Continuous deployment and functional safety have historically been near-mutually exclusive; his research area is precisely the reconciliation the webinar is selling.

What the session actually covers

According to automate.org, the hour maps to five concrete takeaways:

  • Why traditional certification approaches no longer suffice as Physical AI reshapes functional safety
  • How to architect isolated safety domains that simplify AI updates, compliance reassessments, and long-term maintenance
  • How pre-certified technologies and existing safety evidence reduce development effort, audit risk, and time-to-market
  • How Intel and QNX isolate AI, control, and safety functions so teams can update AI workloads while minimizing compliance impact
  • Practical strategies for keeping intelligent robots safe, secure, and supportable as systems evolve

The standards stack, and where the pressure lands

The compliance burden the webinar names is real and multidirectional. ANSI/A3 R15.06-2025 and ISO 10218:2025 modernize industrial robot safety requirements; the EU Cyber Resilience Act pushes security-by-design obligations onto connected products sold in Europe; the Machinery Regulation folds software updates into the safety-law perimeter; and ISO 25785-1 extends coverage to humanoid robots, a category that barely existed when the parent standards were drafted.

A team shipping an AI-powered industrial arm or humanoid in 2027 is, on the organizers’ account, juggling all of these at once — which is why the session treats compliance as an architectural property, not a documentation exercise.

If your roadmap includes shipping AI-driven robots into production, the session’s practical pitch is worth an hour: decouple the certified safety layer from the fast-moving AI layer, lean on pre-certified components, and treat the safety case as a living architecture rather than a certificate on the wall. Registration is open via automate.org’s event listing for the February 11, 2027 session.

Photo of author

Technology Desk

Technology Desk at Archyde.

Chattanooga Nurses at CommonSpirit Memorial Vote to Join NNOC Union