Cyber Agencies Warn of AI-Driven Infrastructure Attacks
A recent Acronis report flags the increasing use of AI by cybercriminals to attack critical infrastructure, including operational technology and SCADA systems. In response, both U.S. CISA and UK's NCSC have issued urgent advisories to infrastructure operators. The alerts emphasize the need for continuous vulnerability management and incident response readiness.
- Threat actors are increasingly using AI not to invent new attacks, but to scale and automate existing methods like phishing and ransomware, with 80% of ransomware-as-a-service vendors advertising AI features. This has led to a 16% year-over-year increase in email-based attacks per organization. - A key tactic is the use of generative AI to create polymorphic malware, which dynamically alters its code to evade signature-based detection tools. Proof-of-concept malware like BlackMamba uses AI APIs to synthesize malicious keylogging code in-memory, making each instance unique and difficult for traditional endpoint detection to identify. - The CISA advisory was prompted by a December 2025 attack on Poland's energy sector, where attackers used compromised VPNs and exploited outdated edge devices to access and damage remote terminal units (RTUs). This highlights the vulnerability of legacy OT hardware, which often prioritizes reliability over security and is difficult to patch without causing downtime. - In 2025, nearly three-quarters of intrusions into OT environments involved the exploitation of remote access points like VPNs. Dragos, a cybersecurity firm, tracked 3,318 ransomware attacks against industrial organizations in 2025, with threat groups like "Azurite" specifically targeting engineering workstations that control physical processes. - Defenders are also leveraging AI for threat detection in OT environments by establishing baseline operational patterns and identifying anomalies that could indicate an attack. However, these defensive AI systems are themselves vulnerable to adversarial attacks, where manipulated input data can cause the AI to misinterpret a threat. - The convergence of IT and OT networks has expanded the attack surface, allowing threat actors to move laterally from enterprise systems to critical control systems. Many industrial environments suffer from poor network segmentation, meaning a single breach can provide an attacker with broad access. - Security experts predict that by mid-2026, a major enterprise will be breached by a fully autonomous AI system capable of executing the entire attack lifecycle from reconnaissance to data exfiltration without human intervention. This is driving the need for security teams to shift from reacting to alerts to proactively hunting for malicious behavior and intent. - Vulnerabilities in edge AI and ML deployments for industrial automation are a significant concern. These resource-constrained environments require efficient model optimization, and the integration of AI with legacy automation systems presents a major challenge for security.
