AI adoption outruns governance across five critical sectors, CISOs report
Agentic systems that plan and execute actions using credentials and connections create a governance gap static certification cannot close.
TL;DR
The Internet Security Alliance surveyed CISOs across defense, energy, health, financial services, and IT sectors and found AI adoption outrunning governance everywhere, including at the most capable organizations. Agentic AI that plans and executes actions using credentials and connections to other systems presents a fundamentally different challenge from the analytic models these sectors have managed before. The CISOs converged on risk-tiered, incentive-based approaches over prescriptive mandates, and all hold that board accountability must rest on evidence rather than assurances. Smaller entities "cannot carry this burden alone."
The ISA assessment, based on CISO reports from five critical-infrastructure sectors and detailed in three blog posts by ISA president Larry Clinton, surfaces a convergence that should worry policymakers: AI adoption is outrunning governance in every sector, and the governance models on the shelf weren't built for what's coming.
The most striking finding is the distinction between two kinds of AI. Analytic and predictive systems, which these sectors have been managing for years, produce output a person then acts on. Agentic AI plans and executes actions on its own, wielding credentials and connections to other systems. That's not an incremental change. It's a category shift, and it demands controls that don't exist yet in most frameworks.
There's a constructive piece, though. The CISOs reported that AI risk sits primarily in the operating envelope around the model rather than in the model itself. That means most of the needed controls are extensions of disciplines critical-infrastructure operators already run, applied to a new category of actor. The problem isn't that the playbook is wrong; it's that nobody's written the AI appendix.
The assessment also identifies a mismatch between static governance and dynamic systems. Annual assessments and point-in-time certifications can't keep pace with models that retrain continuously, acquire new tools, and change behavior between review cycles. Meanwhile, attackers are using AI to operate at speeds human review cycles can't match. For sectors with physical operations, containment options remain coarse enough that isolating a compromised AI component often means interrupting service.
The threat landscape the CISOs described is "industrialized rather than exotic," Clinton writes: fraud at scale, synthetic identity, convincing impersonation, training-data manipulation, and third-party compromise. These aren't hypotheticals. Smaller entities, from community banks to rural hospitals to small defense suppliers, cannot manage them alone.
The policy implications are significant. All five sectors told ISA they prefer risk-tiered, incentive-based approaches over uniform prescriptive mandates. All five hold that accountability belongs at the board level and must rest on evidence, not assurances. And all five, implicitly, are telling policymakers that the current sector-based regulatory architecture, built on industrial-age distinctions, may not survive the technology.
ISA expects to brief Congress on the findings September 2.
Published ·Updated ·Deep Fathom