UK Cyber Chief Warns AI Threats to Critical Infrastructure Will Ramp Up by 2028

The U.K. National Cyber Security Centre (NCSC) has issued a stark assessment regarding the rapid integration of artificial intelligence into adversary playbooks, warning that British critical national infrastructure will face AI cyberattacks by 2028. Speaking at the release of the NCSC’s annual cyber report, CEO Richard Horne warned that hostile state actors—specifically China, Russia, and Iran—alongside cybercriminals are weaponizing machine learning capabilities to scale discovery and exploitation of long-neglected software vulnerabilities across essential services.

The warning underscores a fundamental shift in the threat landscape: while AI is unlikely to instantly invent novel threat vectors out of thin air, it dramatically compresses the time required to weaponize known vulnerability classes and unearth technical debt that organizations have ignored for years.

Accelerated Vulnerability Discovery and Automated Exploitation

Adversaries are actively leveraging AI models to automate reverse-engineering, refine fuzzing engines, and speed up script generation for exploit chains. Rather than creating entirely unprecedented attack vectors, machine learning is primarily amplifying the scale, speed, and efficiency of existing tactics, techniques, and procedures (TTPs).

This trend was highlighted by the recent AI-assisted discovery of two critical vulnerabilities in a primary programming language that underpins approximately 70 percent of global web infrastructure. Those flaws allowed unauthenticated remote actors to gain administrative access to underlying web servers without credentials. When vulnerability discovery is automated at this scale, the traditional grace period between vulnerability disclosure and broad exploitation shrinks from days or weeks to hours.

For organizations running legacy codebases or unpatched perimeter appliances, AI-driven recon allows threat actors to continuously sweep public IP space for specific memory corruption bugs, broken authentication checks, or exposed management interfaces. Once identified, automated agents can deliver customized exploit payloads instantly across thousands of targets simultaneously.

Operational Technology and Supply Chain Blast Radius

The threat to critical infrastructure is particularly acute due to the convergence of corporate IT and operational technology (OT). Many essential sectors—including water treatment, manufacturing, power generation, and cold-chain logistics—rely on legacy Industrial Control Systems (ICS) and Programmable Logic Controllers (PLCs) that lack modern memory protections, centralized authentication, or granular audit logging.

Horne highlighted physical process controls as prime targets for automated disruption, including electricity switching networks, dam sluice gates, and commercial heating and refrigeration units. Recent cyberattacks targeting refrigeration controls across European supermarket chains demonstrated how civilian supply chains can be paralyzed by low-complexity attacks against industrial IoT systems. When automated cooling systems are compromised, entire perishable supply chains fail, leading to immediate public availability issues.

The broader supply chain risk was underscored by the 2025 cyberattack against automaker Jaguar Land Rover, which caused over $260 million in direct financial damage and created ripple effects across downstream suppliers and the wider British economy. Across all sectors, cyber incidents now cost U.K. organizations approximately $20 billion annually.

State-Sponsored Threat Actors: TTPs and Geopolitical Goals

The NCSC report categorizes the primary nation-state drivers behind these escalating campaigns:

• China: Characterized as the U.K.’s most sophisticated and sustained cyber threat. Chinese Advanced Persistent Threat (APT) groups routinely utilize stealthy “living-off-the-land” (LotL) techniques—abusing native administrative tools like PowerShell, WMI, and valid credentials—to maintain persistence within target networks without triggering traditional endpoint detection signatures.
• Russia: Routinely conducts aggressive cyber campaigns designed to hover just below the threshold of open military conflict. Russian intelligence services, specifically the FSB, have repeatedly targeted civil grid operators across Western Europe. Notably, an FSB campaign targeting Poland’s electrical power grid in December 2025 came close to severing power for over 500,000 citizens during peak winter months.
• Iran: Focuses primarily on destructive attacks against commercial entities alongside targeted reconnaissance and surveillance operations aimed at political dissidents and regime opponents residing in the U.K.

As these adversary groups incorporate machine learning into their reconnaissance and command-and-control (C2) frameworks, their ability to evade static indicators of compromise (IOCs) and dynamically obfuscate malicious payloads will increase substantially.

Defending Legacy Systems Against Automated Threats

Addressing this threat shift requires moving away from reactive patching cycles toward proactive surface reduction and strict zero-trust network architectures. Both the NCSC and GCHQ Director Anne Keast-Butler emphasized that fundamental security hygiene remains the most effective defense against AI-scaled attacks:

• Enforce Strict Identity Boundaries: Eliminate public access to operational technology, supervisory control interfaces, and management portals. Require multi-factor authentication (MFA) utilizing phishing-resistant protocols (such as FIDO2/WebAuthn) for all remote access vectors, particularly vendor supply chain maintenance channels.
• Minimize Legacy Technical Debt: Prioritize the inventory and isolation of legacy software stacks. Web-facing systems running vulnerable language interpreters or obsolete frameworks should be placed behind web application firewalls (WAFs) while migration paths are executed.
• Limit Administrator Access: Implement principle-of-least-privilege policies across both Active Directory and Linux environments to block post-exploitation lateral movement. Restrict native administrative tooling on endpoints to prevent living-off-the-land techniques.
• Maintain Immutable Backups: Standardized, isolated offsite backups must be maintained for both IT domain controllers and OT configuration logic files to ensure rapid recovery in the event of widespread ransomware deployment.