Isolating Vital Systems: 2026 Recommendations for Critical Infrastructure, OT, and AI

Introduction

Between April and June 2026, a series of advisories and guidance documents published by CISA and government partners reiterated the need to isolate critical systems. Topics ranged from the persistent risk of Russian intelligence services targeting commercial communications, to hardening OT components, and extending Zero Trust and SBOM principles to artificial intelligence and industrial environments. This article summarizes the key points, providing an updated technical overview for those involved in protecting essential infrastructure.

Technical Analysis

Commercial Messaging Under Advanced Attack: On June 26, 2026, it was highlighted how Russian APTs continue to target consumer messaging applications – email clients, instant messaging, collaborative platforms – to intercept sensitive communications and gain an initial foothold in critical networks. By using "living-off-the-land" techniques, attackers bypass traditional security controls and blend in with legitimate traffic.

ATG Devices Still Exposed: On June 2, a joint fact sheet spotlighted Automatic Tank Gauges (ATGs), automatic tank measurement systems widely deployed in fueling stations, oil depots, and chemical plants. Despite known vulnerabilities, many ATGs remain directly connected to IT networks or the internet, accessible via legacy protocols without robust authentication, creating a direct attack vector toward industrial processes.

AI Supply Chain Transparency: On May 12, the publication Software Bill of Materials for AI – Minimum Elements was released, extending the SBOM concept to AI models and pipelines. With AI increasingly present in critical decision-making systems, the lack of a software component and model bill of materials makes it difficult to track dependencies, inherited vulnerabilities, or backdoors, increasing the risk of cascading compromise.

Zero Trust for OT Environments: On April 29, a specific document adapted Zero Trust principles to Operational Technology, providing practical guidance for microsegmentation of field networks, conditional access, and continuous authentication. The IT/OT convergence has expanded the attack surface, and the absence of Zero Trust models facilitates lateral movement and sabotage of physical processes.

Impact

Insufficient adoption of the recommended measures can cause disruptions of essential services (fuel distribution, energy, water), tampering with pumping systems potentially leading to environmental disasters, compromise of the confidentiality and integrity of critical communications, and alteration of automated AI-driven decisions, with cascading effects on operational safety and public safety.

Mitigation

Key recommendations emerging from the documents are:

• Zero Trust for OT: Implement microsegmentation, least privilege for users and devices, continuous authentication, and granular traffic visibility, taking into account the real-time constraints of industrial environments.
• ATG Hardening: Disable unnecessary network services, promptly apply security patches, segregate devices via firewalls/VLANs, actively monitor logs, and maintain an updated asset inventory.
• SBOM for AI: Require suppliers to provide a complete bill of materials for software components and AI models, integrating it into vulnerability management to swiftly identify vulnerable libraries or models.
• Communications Protection: Adopt end-to-end encryption and dedicated environments for sensitive communications; prohibit the use of consumer applications on critical networks.
• Regulatory Alignment: Refer to CISA guides, NIST SP 800-82, and the Zero Trust for OT framework as a basis for corporate policies.

FAQ

1. What does Zero Trust for OT entail and why is it different from IT? Zero Trust for OT applies the "never trust, always verify" principle to industrial networks, but must adapt to strict constraints: minimal latencies, proprietary protocols, and legacy devices that are not always updatable. It requires microsegmentation, granular authentication, and continuous monitoring, while preserving the reliability and availability of physical processes.

2. What specific risks do exposed ATGs pose? ATGs are often installed without adequate security measures, connected directly to the internet or to IT networks via cellular modems or obsolete serial interfaces. An attacker who compromises them can alter measured levels, disable leak alarms, or cause spills, with environmental, economic, and physical safety consequences.

3. How does SBOM for AI help protect critical systems? An SBOM provides a transparent list of all software components and AI models used, including dependencies. This allows verifying the absence of known vulnerabilities, ensuring supply chain integrity, and rapidly responding to new threats, reducing the risk that a compromised model influences safety-critical decisions.