Tue Nov 03
Snapshot Compliance Meets Continuous Agentic Drift
NERC CIP and IEC 62443 both certify a moment in time, but agentic OT controllers change behavior between audits, leaving buyers with no framework for continuous assurance.
Certification activity around IEC 62443 is accelerating. Abbott Risk Consulting has joined ISASecure to expand assessment capacity, ODVA’s CIP Security has earned IEC 62443-4-2 SL-C 3 certification for EtherNet/IP products, and Fortinet Korea is bringing CISOs and plant engineers together in Seoul to walk through implementation specifics as AI widens the attack surface in manufacturing and energy.
What gets less attention is that a second, mandatory regime governs a large share of this same hardware once it sits inside the power sector. NERC’s Critical Infrastructure Protection standards set the cybersecurity requirements for bulk electric system assets, and utilities running IEC 62443-certified components on the grid are also on the hook for CIP compliance. Buyers evaluating AI-enabled automation now have two assurance regimes layered on the same equipment, and both share a structural weakness that agentic systems are exposing.
Snapshot assurance, continuous behavior
IEC 62443 certification and NERC CIP compliance are both demonstrated at a point in time: a component passes conformance testing, a utility passes its audit, and the credential holds until the next cycle. That model works when the underlying system is static. It breaks down when the control layer is an agent that reasons and adjusts in production, because nothing in either regime re-verifies behavior between certification events.
The pressure to deploy that kind of control layer is building faster than the assurance model is adapting. Neuron Industries’ $10 million seed round is explicitly built to compress automation development from months to days. Faraday’s new EtherCAT-compliant Ethernet PHY IP can synchronize factory machines to within a tenth of a millionth of a second, purpose-built for AI-driven manufacturing and humanoid robotics. Faster deployment and tighter coordination both shrink the window during which a certified snapshot remains an accurate picture of what the system is actually doing.
Credentialing people is easier than credentialing behavior
Workforce credentialing is moving in response. EC-Council has expanded its AI certifications to build trusted readiness among the engineers who will operate these systems. That is useful, but it addresses who is qualified to supervise, not whether the agent’s decisions between audits can be independently verified.
A model for what that verification could look like already exists outside OT. Cyara’s agentic CX trust platform produces independent proof of quality and compliance on every live agent interaction, not a periodic sample. Nothing comparable exists yet for an agent controlling a turbine or a production line governed by NERC CIP and IEC 62443.
What this means for the compliance decision
A 62443 certificate and a clean CIP audit are both necessary, and neither was designed to answer whether an agentic control layer is behaving safely right now, between cycles. Utilities and industrial operators deploying agentic automation should treat both credentials as floors and ask vendors a separate question directly: what continuous, per-decision evidence exists that the agent’s behavior today matches what was certified at commissioning. Until that evidence exists as a standard deliverable, the gap between certified and currently safe is the operator’s exposure to carry, audit cycle by audit cycle.
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 point-in-time certification regimes create a structural gap when applied to continuously adapting agentic systems—is coherent and defensible, though the Cyara analogy (customer |
| Source & Claim Verification | In-house seat · hosted fallback (qwen-2.5-72b-instruct) | cleared. All claims are supported by the sources that could be opened, but some sources were unverified due to access issues. |
| Regulatory & Framework Fidelity | Mistral | cleared. The briefing accurately addresses NERC CIP and IEC 62443 but omits critical alignment with ISO 42001, EU AI Act, NIST AI RMF, and sector-specific frameworks (HIPAA, CLIA, DO-178C, etc.), which are man |
| Technical Accuracy | Llama | cleared. The article accurately describes the limitations of current certification regimes (IEC 62443 and NERC CIP) in addressing the dynamic nature of agentic systems in industrial automation. |
| Bias, Balance & Hype Control | Gemini | cleared. The briefing effectively identifies a critical gap in existing compliance frameworks when confronted with continuously adapting agentic systems, and it largely avoids vendor hype by focusing on the st |
| Novelty & Non-Duplication | Grok | held. Core thesis—point-in-time IEC 62443/NERC CIP credentials vs continuous agentic exposure—is a near-restatement of the 2026-10-23 briefing ‘Industrial AI Certification Is a Snapshot. The Exposure Is Con |
| Validation | DeepSeek | cleared. The central claim that IEC 62443 and NERC CIP rely on point-in-time ‘snapshot’ assurance, creating a gap for continuously learning agentic systems, is validated by the standards’ own audit-based, peri |
Sources cited: 15. Validation challenges: 0. Learn how these briefings are written and verified.