Why Critical Infrastructure Needs Zero-Trust Security

Malicious cyber activity targeted remote monitoring and control technology at over 30 community water systems across Minnesota in late July 2026, and water cyber attacks were subsequently reported across at least 12 states. However, the underlying OT security weaknesses are not exclusive to the water sector. Energy utilities, manufacturing plants, transportation systems and public sector facilities run the same types of Operational Technology (OT) and rely on similar third-party integrators for maintenance, using equipment installed before anyone thought it could be reached from the internet. Critical infrastructure needs zero-trust security because these environments still grant access based on where a user or device resides on the network, and attackers are now exploiting that assumption.

Continue reading to learn how traditional security models have become insufficient, why AI expedites the speed at which OT weaknesses can be found by attackers and how to apply zero trust to OT environments to protect critical infrastructure.

How the traditional security model fails in critical infrastructure

Internet-exposed controllers, default credentials and always-on vendor access are known weaknesses, and none require sophisticated technology or expertise to exploit because they rely on trust assumptions. A flat or lightly segmented network assumes that anything already inside belongs; an engineering workstation with reach into all controllers assumes the person at the keyboard is who should be there. Similarly, a default password assumes the device sits somewhere an outsider cannot reach.

The U.S. water cyber attack exposed a quieter version of the same underlying issue: Some reachable equipment consists of cellular modems and devices installed by vendors that are never captured in routine security scans. Most of these environments were designed under conditions where every assumption was made at installation and inherited by every team since then without ever being re-examined. Nobody intends to leave a PLC on the public internet, but IT/OT convergence has moved the boundary in recent years, and the trust map hasn’t been adjusted to match.

Why AI shortens the time you have to fix OT weaknesses

In joint advisory AA26-231A, issued on August 19, 2026, the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency reported that cybercriminals are using AI-generated scripts, disguised as legitimate monitoring utilities, to conduct reconnaissance against internet-exposed Siemens S7 Series PLCs. The agencies warned that AI-assisted development can lower the technical barrier and reduce the time required to create exploitation tooling. The advisory does not describe new zero-day vulnerabilities, and the activity it observed is reconnaissance rather than confirmed disruption. The weaknesses are the same ones already in federal advisories, but what’s changed is who can act on them and how quickly.

AI-driven security concerns are also extending to the identity side. On August 27, 2026, more than 100 organizations signed an open letter on collective cyber defense warning readers that AI-enabled cyber attacks could become more widespread soon, naming hospitals, water treatment plants and internet infrastructure among the systems at risk and calling for support for under-resourced critical infrastructure defenders.

How zero trust for critical infrastructure changes OT security

Zero trust is a security model that removes implicit trust; instead of granting access based on network location, it verifies every user and device by identity, limits access to the minimum required and logs activity. In OT environments, zero trust means being on the control network no longer entitles anyone to reach the equipment on it. Now, there is a shift in what needs to be defended, which is why zero trust works in environments where the boundary has already dissolved, and the equipment behind it cannot be replaced.

Identity becomes the control point, not network location

Every human, vendor, service account and AI agent that touches your environment gets a verified identity, and access decisions are made against that identity rather than an IP address. A shared operator login fails this test instantly because it identifies a role, not a person. The AI industry has put the same standard on record for its own systems. The collective letter states that agentic identities should be traceable and accountable — a commitment from the companies building these agents, not a directive to infrastructure operators. It’s still worth holding your AI vendors to that standard and applying it to any agent running in your environment to ensure AI agents do not have standing access.

Least privilege and JIT access are the priority, not standing access

Consider an integrator who services a treatment plant every six months and needs several hours of access each time. Under a traditional cybersecurity model, that vendor holds a permanent VPN account, which means their credentials remain a live path into your network for the rest of the year, protected only by their internal security practices. Just-in-Time (JIT) access inverts that; the account is provisioned only when the work is scheduled, scoped to the specific system and revoked automatically when the window closes. Zero-Trust Network Access (ZTNA) enforces the same principle for the connection itself; instead of placing the vendor on your network like a VPN does, it connects them to one system and makes nothing else visible.

Limit the blast radius across the IT/OT boundary

Containment assumes someone will eventually gain access and questions how far they can go with that access. Segmenting the IT/OT boundary, restricting which engineering workstations may communicate with which controllers and eliminating standing access all minimize the blast radius. The same logic applies to AI agents operating across your systems, where permissions accumulate quickly and quietly. Try our AI agent blast radius calculator to see how far a single compromised AI agent could reach.

Continuously verify, monitor activity and record sessions

Login verification requires one check, but continuous verification re-evaluates access as conditions change and revokes access when they do. Session monitoring and recording capture exactly what happened during a privileged connection to a controller, turning a suspicious event into something you can investigate rather than reconstruct after the fact.

How to apply zero trust to OT environments

In OT environments, the main security priorities are availability and safety because a control system going down means people may not get water or power. CISA’s guidance on adapting zero-trust principles to OT maps zero-trust outcomes to six NIST CSF 2.0 functions while clearly accounting for legacy gaps, operational constraints and safety requirements. The guidance recognizes that zero-trust principles must be adapted to OT environments rather than applied identically to traditional IT systems. That’s the practical case for zero trust in OT. Advisory AA26-231A also tells operators to use compensating controls when a legacy device cannot be patched or reconfigured without a planned outage, and the collective letter makes the case for the same strategy. An outdated controller may not be able to be patched, but who reaches it, from where and for how long can be controlled and monitored. Here’s where to start when applying zero trust to OT environments:

Protect your critical infrastructure with Keeper

Recent cyber attacks and federal advisories show why critical infrastructure organizations need stronger controls around privileged and remote access, particularly when legacy OT cannot be quickly patched or replaced. Organizations can reduce risk by controlling who can access critical systems, what resources they can reach, how long access lasts and what actions they can take. KeeperPAM® helps critical infrastructure organizations secure and manage privileged access across IT and OT environments. Through the Keeper Gateway, it provides secure remote access to infrastructure without opening inbound firewall ports or exposing credentials to end users. Built on a zero-trust and zero-knowledge architecture, KeeperPAM also supports JIT access, automated credential rotation, least-privilege access and privileged session management. Start a free trial of KeeperPAM to close the access paths attackers are already scanning for.