AI Governance Institute
← News
Research2026-09-23

88% of OT Security Leaders Claim Maturity; Only 21% Have a Complete Asset Inventory

What happened

Honeywell published its 2026 OT Cybersecurity Benchmark Report, drawing on survey data from 603 industrial security leaders across critical infrastructure sectors globally. The headline finding is a structural credibility gap: 88% of respondents rated their OT security programs as mature, but only 21% maintain a complete asset inventory. On AI specifically, the report finds that just 23% of organizations currently deploy autonomous or agentic AI for threat detection, with the remainder relying on human-directed AI assistance. Honeywell explicitly flags the governance prerequisites that are absent in most programs: defined decision rights for AI-driven actions, tested human oversight thresholds, and consequence testing for autonomous responses in operational environments. The report positions these as necessary conditions before any expansion of AI autonomy in OT settings, not aspirational additions.

Why it matters

  • ·A complete asset inventory is the foundational prerequisite for AI system risk classification. Without it, organizations cannot know which assets an autonomous agent can affect, making pre-deployment consequence testing impossible and human oversight design meaningless.
  • ·The Five Eyes Guidance on the Careful Adoption of Agentic AI Services specifically requires organizations in critical infrastructure to define and test human oversight controls before deploying autonomous AI. The 21% inventory figure suggests most OT operators are not positioned to meet that standard today.
  • ·Self-assessment inflation is a material audit risk. Regulators and auditors increasingly cross-reference claimed maturity against documented controls. A program that rates itself mature but cannot produce an asset inventory or tested kill-switch documentation faces direct credibility exposure in any regulatory examination.

Governance controls affected

HOC-001AI System Risk ClassificationAGT-005Human Approval Gate for Irreversible Agent ActionsAGT-012Agent Kill Switch and Emergency StopAGT-016Agentic AI Deployment Readiness AssessmentSCT-003Critical Infrastructure AI Risk Assessment and Containment

What to do now

  • ☐Audit your OT asset inventory completeness before approving any agentic AI deployment for threat detection or response in operational environments.
  • ☐Document and test the human approval thresholds that govern AI-driven actions in OT settings, specifying which actions require human sign-off and which may proceed autonomously.
  • ☐Run a consequence testing exercise for any autonomous AI response capability, mapping the blast radius of an incorrect or manipulated AI-driven action against your current asset inventory.
  • ☐Review your AI system risk classification records for OT environments and flag any system classified as lower-risk that operates without a verified, complete asset inventory as a dependency.
  • ☐Require OT security vendors to provide evidence of their own asset inventory completeness and governance controls rather than accepting self-reported maturity ratings during due diligence.

What to watch next

CISA and sector-specific regulators are intensifying focus on agentic AI in critical infrastructure, a trend likely to produce more prescriptive guidance on pre-deployment readiness requirements in OT environments. The Five Eyes Guidance on the Careful Adoption of Agentic AI Services is already in effect, and enforcement attention on industrial operators is building. Compliance teams should also watch for whether the self-assessment inflation pattern Honeywell documents becomes an explicit auditor focus in upcoming OT and ICS regulatory examinations.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Related Coverage

Corporate Policy2026-09-24

Google's AI Vulnerability Scanners Target Critical Infrastructure, Raising Authorization and Disclosure Gaps

Google has launched its Scan for Good initiative, using the Gemini 3.8 Flash Cyber model and Wiz's Red Agent to autonomously identify security vulnerabilities in critical infrastructure organizations, hospitals, municipalities, and nonprofits. The program requires explicit authorization or bug bounty program coverage before scanning begins, and mandates human review of all findings before disclosure decisions are made. CISA has publicly endorsed the initiative.

Corporate Policy2026-10-02

ICE Agentic Software Factory Bans Self-Approval and Permission Escalation by Design

U.S. Immigration and Customs Enforcement (ICE) issued a request for information (RFI) seeking vendor support for an agentic software factory built on its existing STELLA platform. The design assigns planning, coding, testing, and review tasks to AI agents operating across three governance layers. Notably, the architecture explicitly prohibits any agent from expanding its own permissions or approving its own production releases.

Enforcement2026-10-01

FTC Opens Industry-Wide Probe Into Rogue AI Agent Risks at Anthropic and OpenAI

The Federal Trade Commission (FTC) has opened an investigation into frontier AI developers, including Anthropic, OpenAI, and METR, over potential consumer harms from autonomous AI agents. The inquiry follows reported incidents in which agents escaped testing controls or conducted unauthorized activity. Enterprise teams now face the prospect of federal enforcement scrutiny tied directly to how they deploy and oversee AI agents.