Agentic AI Is Outrunning Machine Safety Certification

Agentic AI Is Outrunning Machine Safety Certification

The machine safety market is projected to reach $16.09 billion by 2035, with AI-native entrants now sitting alongside the automation conglomerates that have defined the category for decades, according to EIN Presswire. Buyers already understand that a system flagging a pinch-point risk is not the same as a system certified to stop the machine when that risk materializes. The line between AI-adjacent analytics and IEC 61508 or ISO 13849 certified safety functions is well established. That is not the hard part anymore.

The hard part is what agentic deployment does to that line. Industrial AI is moving from advisory tools that flag anomalies to systems that act on them, a shift iot-analytics’ mid-2026 industrial AI pulse check frames as this year’s defining deployment question. What that pulse check does not resolve is a compliance question sitting directly underneath it: an agent that adjusts a setpoint or throttles a process without a human in the loop is, functionally, a safety-relevant component. Under the EU AI Act’s risk-tier structure, AI embedded in machinery safety functions is squarely in high-risk territory, yet SIL and PL classifications under IEC 61508 and ISO 13849 were architected for deterministic control logic, not for systems whose behavior is learned and can drift.

Vendors have not closed that gap by citing cybersecurity standards. The ISA/IEC 62443 series is frequently invoked as covering AI-introduced vulnerabilities in automation systems, but as ANSI’s analysis of AI risk assessments for critical infrastructure makes clear, that series secures control systems against intrusion. It says nothing about what happens when an autonomous agent acts correctly by cybersecurity standards and wrongly by functional safety standards.

Tooling is starting to fill the evidentiary gap, though not inside the certification bodies that would need to act on it. Red Hat’s asago project, built with IBM Research, NVIDIA, Microsoft, and MIT Lincoln Laboratory, generates structured, verifiable evidence of how an AI system was built and governed, according to HPCWire and AI Magazine. That kind of provenance trail is exactly what ISO 42001 audits and EU AI Act technical documentation obligations will demand, but it is a governance artifact, not a substitute for a SIL rating.

The same gap is opening in a second regulated vertical. Dassault Systèmes’ planned acquisition of ArisGlobal to add AI and compliance capabilities into life sciences workflows, reported by ARC Advisory Group, bundles autonomous AI action into a workflow that FDA AI/ML software guidance and MDR/IVDR conformity assessment were built around human-reviewed decisions, not agents acting inside the process.

The near-term task for buyers has shifted with the technology. It is no longer sufficient to confirm that a vendor’s AI sits outside the certified safety function, or outside the reviewed clinical workflow. Buyers need to ask which risk tier the agent’s autonomous action falls into under the applicable framework, whether that tier has been formally assessed, and whether the governance evidence a vendor produces was built to satisfy an auditor or a regulator. Those are not the same document, and right now too many vendors are selling the former as if it were the latter.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core argument—that agentic AI creates a regulatory gap between existing deterministic safety certifications (SIL/PL) and AI governance frameworks—is logically coherent and well-supported, though t
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but some sources could be more robust or directly relevant to the claims they support.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects the regulatory gaps between agentic AI and existing frameworks (ISO 42001, EU AI Act, FDA, MDR/IVDR), but lacks explicit citations of specific clauses or annexes from
Technical Accuracy	Llama	cleared. The article accurately highlights the challenges of applying traditional safety certification standards to agentic AI systems, citing relevant regulations and industry developments.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and critiques vendor hype by distinguishing between governance artifacts and formal safety certifications, highlighting a critical gap in current offerings.
Novelty & Non-Duplication	Grok	cleared. The core thesis—that agentic closed-loop action collides with deterministic SIL/PL regimes and that governance provenance is being sold as if it were functional-safety evidence—is a real synthesis, no
Validation	DeepSeek	cleared. The central claim that agentic AI is creating a compliance gap with established safety certification frameworks is validated by expert analyses and vendor actions, but the briefing does not provide a

Sources cited: 12. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.