77% of IT Managers Say AI Agents Are Out of Control. Here's What That Means.
Seventy-seven percent. That is the share of IT managers who now say AI agents in their organizations are out of control. Not "difficult to manage." Not "a work in progress." Out of control.
That number, surfaced in a ZDNet industry survey, is the headline. But the supporting data makes it worse. According to the Cloud Security Alliance, surveying 418 enterprise respondents in January 2026, 82% of enterprises have unknown AI agents running in their IT infrastructure right now. Not agents they haven't gotten around to auditing. Unknown agents. Operating without sanction, without policy, without any accountability chain.
Sixty-five percent of those same enterprises have already experienced an AI agent-related incident in the past year: data exposure in 61% of cases, operational disruption in 43%, direct financial cost in 35%. These are not theoretical future risks. They are current operational realities being reported by enterprises that have already absorbed the impact.
The Setup: Deployment Without Governance
How did we get here? The Monte Carlo survey of 260 enterprise leaders puts it plainly: 64% deployed AI agents before feeling fully prepared. That is not a technology failure. It is a governance failure created by the gap between deployment velocity and the infrastructure required to govern what gets deployed.
The same survey found that only 47% of builders say their systems are easily traceable end-to-end when something goes wrong. More than half of the people building the systems that enterprises depend on cannot follow the decision trail when an agent makes a mistake. That is not an edge case. That is the default state of enterprise AI today.
Deloitte's AI Institute 2026 survey adds the strategic dimension: 74% of companies plan to deploy agentic AI within two years. Yet only 21% have a mature model for AI governance. The deployment wave is accelerating. The governance infrastructure is not keeping pace.
What "Out of Control" Actually Means
"Out of control" is not hyperbole. It is a technically precise description of the default state of enterprise AI agent deployments. An agent that is "out of control" has four specific deficits:
No cryptographic identity. The agent has no verified, attestable identity. You cannot prove which agent took which action, because there is no signature on the action. You have logs, maybe. You have records, probably. But you do not have cryptographic proof — the kind that survives legal discovery, regulatory examination, or a forensic investigation after something goes wrong.
No policy enforcement at execution time. Policies exist in documents. Governance frameworks exist in slide decks. But when the agent calls the API — when it executes the action — there is no enforcement mechanism operating between model intent and execution. The agent acts, and the audit happens afterward, if at all.
No immutable audit trail. Logs are not audit trails. Logs can be deleted, manipulated, or simply not generated. An immutable audit trail is a cryptographically signed, tamper-evident record of every agent action, created at execution time, anchored in a structure that cannot be retroactively modified. Most enterprises do not have this for their AI agents.
No incident response capability. Grant Thornton surveyed 950 business leaders between February and March 2026 and found that just 20% have a tested AI incident response plan for when agents fail. When something goes wrong — and the data says it will — four out of five enterprises have no rehearsed response. They will improvise, under pressure, with incomplete information, while regulators and counsel are waiting.
The Regulatory Consequence
Every regulated industry is now one incident away from an enforcement action.
Healthcare organizations deploying clinical AI agents are operating under HIPAA and emerging FDA clinical AI guidance. Financial institutions deploying AI agents for trading, fraud detection, or customer decisions are operating under FINRA, SEC, and FinCEN frameworks. Industrial operators deploying AI agents in critical infrastructure are operating under IEC 62443 and NERC CIP. Government agencies are operating under NSM-10 and CNSA 2.0 post-quantum mandates.
None of those frameworks were written for autonomous AI agents. But they apply to the decisions those agents make. And when an enforcement action comes — when a patient is harmed by an unaudited clinical AI action, when a financial institution cannot reconstruct an autonomous trading decision, when a critical infrastructure operator cannot prove what their AI did during an incident — the absence of governance infrastructure is not a mitigating factor. It is the liability.
The agentic AI security incident at 90+ organizations — where autonomous agents gained write access to firewalls and IAM policies — is the preview of what uncontrolled AI agent access looks like at enterprise scale. Not theoretical. Reported.
The Infrastructure That Was Missing
CoreIdentity was built for exactly this moment. Not for the moment when AI governance becomes a regulatory requirement. For the moment before that — when the governance infrastructure that prevents enforcement actions can still be put in place.
We are not a framework. We are not a checklist. We are not an advisory practice that produces a roadmap for what you should eventually implement. We are the enforcement infrastructure — operating at execution time, before the API call, before the action, before the record that cannot be changed.
Every agent that runs through CoreIdentity has a cryptographic identity. Every action it takes is evaluated against policy before execution. Every decision is recorded in an immutable audit trail, signed with ML-DSA-65 (FIPS 204) — the NIST post-quantum standard that ensures those records remain verifiable even as classical cryptography becomes vulnerable. We are the only governance infrastructure platform implementing all three NIST FIPS post-quantum standards (FIPS 203, 204, 205) simultaneously in production.
The 77% of enterprises whose IT managers say their agents are out of control are not failing at AI. They are operating without the infrastructure layer that makes AI controllable. That infrastructure layer now exists.
If Your Agents Are in That 77%
The data is not ambiguous. The governance gap is not closing on its own. Every week that autonomous agents operate without cryptographic identity, without policy enforcement at execution time, and without immutable audit trails is a week of accumulated liability.
If your organization is in the 77%, or the 82%, or the 80% without a tested incident response plan — let's talk. Not about a framework. About the infrastructure.
