AI models successfully plan cyberattacks on critical infrastructure in 90% of attempts, Frenos study finds

Research released by cybersecurity company Frenos on October 21, 2026 reveals that leading artificial intelligence models successfully planned cyberattacks against critical infrastructure in approximately 90% of attempts. Across nearly 17,000 tests on industrial systems, top commercial models produced workable attack blueprints, while open-weight models varied widely from 5.2% to 36.2%. Security experts warn that AI accelerates reconnaissance and lowers technical barriers for attackers, compounding vulnerabilities for resource-constrained power and water operators.

Frenos study methodology and overall results

• ▪Frenos researchers tested six frontier models, two U.S. open-weight models, and two foreign open-weight models by providing them target details and asset information to assemble attack steps, without disclosing model names.
• ▪Cybersecurity company Frenos published research on October 21, 2026 showing top artificial intelligence models successfully produced workable cyberattack plans against industrial systems in 89.5% to 92.9% of planning attempts.
• ▪The Frenos study evaluated 10 artificial intelligence models across 16,980 total tests conducted between October 2025 and September 2026, including 13,884 decision attempts, 2,520 planning attempts, and 576 simulated attack runs.
• ▪Frenos emphasized that its cyberattack tests used supplied information and simulated scenarios without proving that artificial intelligence models could independently identify targets or execute cyberattacks.

Performance variations across tested AI models

• ▪In the Frenos study, the strongest foreign open-weight model achieved a 36.2% success rate in attack planning, outperforming the top U.S. open-weight model at 5.2% and three U.S. frontier models scoring between 17.9% and 31.9%.
• ▪The three leading frontier models tested by Frenos produced attack plans that passed every repeat test in 60%, 62.9%, and 74.3% of attempts, showing inconsistency and omitting essential steps or assuming unestablished access.
• ▪Single decision accuracy varied across models in the Frenos study, with models correctly matching required outcomes in 6% to 47% of single action choices.

Expert commentary on AI threat capabilities

• ▪Harry Thomas, co-founder and chief technology officer of Frenos, stated that artificial intelligence allows threat actors to map industrial networks without human time constraints like rest or sleep, accelerating potential attack timelines.
• ▪Steve Swick, former security executive at American Electric Power, stated that utility security historically relied on obscurity, but artificial intelligence is rapidly eliminating the specialized technical knowledge previously required to target infrastructure.
• ▪Paul Shaver, head of operational technology at Armadin, stated that artificial intelligence agents can parse automation controller manuals and conduct reconnaissance on plant design features without triggering security alerts.

Real-world cyberattacks on industrial infrastructure

• ▪In a January 2026 cyberattack on a Mexican water utility, attackers utilized Anthropic's Claude artificial intelligence to identify an industrial monitoring interface, though additional access attempts failed and no operational systems were compromised.
• ▪In a recent incident in France, hackers manipulated supermarket refrigeration systems, causing temperatures to rise sharply, spoiling food, and triggering reports of fires.

Threats to critical infrastructure

• ▪Industrial systems like water treatment and manufacturing plants face elevated risks because many operators face staffing constraints, and unexpected network traffic during attacks can disrupt essential operations.
• ▪Nation-state adversaries and cybercriminals have long targeted critical infrastructure facilities, including water treatment systems and manufacturing plants.