Critical infrastructure operators across the United States face a policy overhaul after the Operational Technology Cybersecurity Coalition published 18 recommendations this week urging Congress and the Administration to fund operational resilience and govern AI agents as AI-enabled attacks on industrial systems accelerate, Industrial Cyber reported.
The coalition’s report, titled “The Missing Pillar of AI Security: Critical Infrastructure Resilience,” argues that national AI security policy has overlooked the systems that actually keep the lights on: the operational technology — programmable controllers, sensors, and SCADA networks — running energy, water, and manufacturing plants. The recommendations were built from four summer roundtables with nearly 70 representatives from critical infrastructure sectors, government, and the cybersecurity community.
The timing is not academic. In August, CISA and federal partners warned that attackers are using AI-generated scripts to target industrial controllers, letting adversaries find and exploit long-known vulnerabilities faster and at greater scale than before. And last week’s nationwide FAA ground stop and Microsoft outage — neither a cyberattack, the coalition noted — briefly stoked fears of a coordinated strike, while demonstrating the scale of disruption adversaries are trying to engineer.
“Congress and the Administration don’t need to start from scratch,” said Michael Garcia, policy director of OTCC. “Every recommendation in this report builds on authorities, programs, and institutions that already exist. What’s missing is a decision to treat operational resilience as part of AI security policy, and to fund it that way.”
Why patching alone can’t save the grid
The report’s hardest-hitting number is this: roughly 70% of OT vulnerabilities sit in legacy or deeply embedded systems that cannot be patched or rebooted without interrupting continuous physical operations. A water pump that runs for months between maintenance windows cannot take a Tuesday-night patch cycle. Meanwhile, CISA tracks more than 12,000 industrial control system vulnerabilities, and a shortage of specialized engineering talent creates severe bottlenecks in testing them.
That math is why the coalition pivots to an “assume compromise” mindset. If most of the attack surface is unpatchable and AI is compressing the time between discovery and exploitation, defense has to shift from prevention to withstand, contain, and recover. The report ties this to existing U.S. strategy language, including NSM-22 and the 2026 National Resilience Strategy, as well as recent guidance from the United Kingdom.
The unglamorous basics AI exposes
Here’s the uncomfortable part, and the report doesn’t sugarcoat it: attackers are still getting in through the same old doors. Weak passwords, missing multi-factor authentication, overprivileged access, poor network segmentation. AI makes those gaps cheaper to find, not harder to close.
The coalition’s answer is a national framework anchored on CISA’s Cross-Sector Cybersecurity Performance Goals — a common security floor, with sector-specific regulators layering on risk-based requirements. The package includes secure-by-design mandates for hardware and software producers, a cyber duty of care, whistleblower protections, compensating controls for unpatchable legacy systems, and financial assistance for under-resourced entities like smaller water utilities.
AI agents become the insider threat
The report’s most forward-leaning section deals with non-human identities. Organizations are deploying AI agents to fight AI-enabled attacks, but adopting them faster than they can govern them. In some environments, NHIs now outnumber human employees by more than 100 to 1, per the report — which means the definition of “insider threat” is quietly expanding to include autonomous software with privileged credentials.
The fix, OTCC argues, is to extend existing controls — zero trust and identity access management — to AI models and agents, treating them like any other privileged identity rather than a new category demanding new bureaucracy. The coalition also calls for an OT-specific AI governance framework, testing of autonomous AI deployments, and long-term access to AI-enabled defensive tools.
The adoption gap on the defender side
Defenders are lagging. Nozomi Networks survey data cited in the report shows fewer than 8% of organizations have deployed AI across multiple OT cybersecurity functions. Attackers, by contrast, need only one good model and one exploit script.
To close that gap, the coalition recommends connective tissue between AI labs, federal and state agencies, industrial automation providers, and operators — including controlled test beds where models can train on sensitive infrastructure data, such as electric utility data kept within owner environments. Federal agencies, it says, should turn pilot findings into practical procurement guidance.
Recovery gets a budget line
The resilience pillar is where money meets policy. The report urges funding the Cyber Response and Recovery Fund, reconstituting the Cyber Safety Review Board, improving coordination between cybersecurity professionals and emergency responders, and adopting CISA’s Critical Infrastructure Fortify controls. Operators should regularly assess whether their teams can actually detect incidents across operational assets and restore priority functions with minimal disruption — ideally through voluntary exercises in realistic conditions.
“A software glitch is an inconvenience, but a deliberate, AI-driven strike on the operational technology powering our essential services would cause untold harm to Americans around the country and impact our national security.”
That warning, from the coalition’s statement following last week’s FAA ground stop and Microsoft outage, frames the entire report: the two incidents showed the blast radius; AI would add speed and targeting.
The 30-second read for security teams
- Baseline controls — MFA, segmentation, access management — remain the decisive factor; AI doesn’t excuse their absence.
- Assume compromise: prioritize detection and rapid restoration of priority functions over perfect prevention.
- Govern your AI agents like privileged identities under zero trust, or they become your largest insider-threat surface.
- Expect harmonized, CPG-based requirements across critical infrastructure sectors, with sector regulators adding risk-based layers.
OTCC plans to convene policymakers, AI developers, and the OT community in the coming months to test and refine the proposals, while pressing Congress and federal, state, and local governments — along with AI laboratories — to implement them. The report’s closing observation is blunt: AI-enabled threats to OT environments are no longer hypothetical.