AI Can't Secure What Was Never Inventoried

AI Can’t Secure What Was Never Inventoried

Manufacturers are being sold AI-driven OT security platforms on the strength of their detection models. The harder question buyers should be asking is whether those models can even see the assets they’re supposed to protect.

ARC Advisory Group’s read on Palo Alto Networks’ growth in industrial cybersecurity makes the constraint explicit: AI is most consequential in OT precisely because organizations must secure large populations of unmanaged assets with limited specialist resources, and Precision AI’s asset discovery and behavioral analysis functions exist to close that gap, not to add a smarter alarm on top of an already-known inventory (ARC Advisory Group). If a plant’s legacy PLCs, sensors, and gateways were never mapped, an AI layer trained to spot anomalies has no baseline to compare against. It will miss what it never learned existed.

This is not a hypothetical gap. Research on IT/OT integration points, cited by Rockwell Automation, identifies those integration seams as the most vulnerable part of industrial networks, the exact terrain where unmanaged and shadow assets tend to live (advancedmanufacturing.org). Rockwell’s position is that resilience requires an end-to-end approach aligned to recognized frameworks, not point tools layered onto an incomplete map.

For regulated operators, the compliance framing sharpens the stakes further. Nozomi’s Compass platform is being positioned specifically to give AI-driven tools contextual OT data that satisfies NERC CIP requirements (Industrial Cyber), and NERC CIP-005 sets explicit electronic security perimeter and access point requirements that recent protocol-aware enforcement research for IEC-104 environments is built to satisfy (ScienceDirect). An AI tool that cannot demonstrate it is reasoning over a complete, current asset and perimeter picture is not compliance-ready under CIP-005, regardless of how well its anomaly detection scores in a vendor demo.

The remedy is not a better model. It is a risk quantification discipline that forces the visibility question before procurement, not after. DeNexus’s work with the ISA Global Cybersecurity Alliance is aimed at exactly this: using the ISA/IEC 62443 framework to assess, quantify, and assign a cost to OT risk, so organizations can see where unmanaged exposure actually sits on the balance sheet rather than trusting a vendor’s coverage claims (Automation.com). ISAGCA’s broader standardization push around IEC 62443 exists for the same reason, giving buyers a shared reference point for what “covered” actually means (Automation.com).

The decision in front of manufacturing and utility buyers this quarter is not which AI security platform has the best detection accuracy. It is whether that platform can prove, against an IEC 62443 zone and conduit model, that the assets feeding its AI are the assets that actually exist on the floor. Skip that step and the AI is only as good as the blind spots it inherited.

Visibility is the compliance artifact. Detection is just what happens after.

---

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 AI security tools cannot protect assets that were never inventoried—is logically sound and well-supported, though the claim that an AI layer ‘will miss what it never learned exi
Source & Claim Verification	Qwen · hosted fallback (qwen-2.5-72b-instruct)	cleared. All factual claims are supported by citations, but some sources (e.g., Instagram, Facebook) may not be the most reliable for academic or professional verification.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects IEC 62443 and NERC CIP requirements but lacks explicit mapping to ISO 42001, the EU AI Act, NIST AI RMF, 21 CFR Part 11, or EU GMP Annex 11/22.
Technical Accuracy	Llama	cleared. The article is technically accurate and correctly conveys the importance of asset visibility in OT security, citing relevant industry frameworks and research.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively counters vendor hype by consistently emphasizing the foundational requirement of asset inventory before AI-driven security can be effective, using credible sources to support
Novelty & Non-Duplication	Grok	held. The thesis that AI detection is worthless without prior OT asset inventory is decades-old conventional wisdom (‘you can’t protect what you can’t see’) merely re-skinned with current Palo Alto/Nozomi/D
Validation	DeepSeek	cleared. The central claim that AI security is ineffective without a complete asset inventory is strongly supported by expert analysis and compliance requirements, which validate it against operational reality

Sources cited: 18. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.

Record corrected 29 September 2026. This seat was originally listed as running on our own local hardware. It never did: every review before this date fell back to a hosted model, and the row above now names the model that actually ran.