AI Models Show High Success in Planning Cyberattacks
Researchers found that top-performing models could generate attack sequences for industrial systems in 90% of attempts.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Do you trust that current security measures can protect our power and water systems from AI-driven threats?
Frenos researchers conducted nearly 17,000 tests on 10 AI models to assess their ability to plan cyberattacks against industrial systems. Three of the models successfully constructed attack plans in 90% of their attempts, according to research released in October 2026.
Why it matters
As industrial operators struggle with budget and staffing constraints, the potential for freely available AI to lower the barrier for executing complex cyberattacks introduces new security risks to essential infrastructure.
Researchers performed 2,520 attempts to construct complete attack plans across 10 models. Frontier models showed a success rate between 17.9% and 31.9%, while three unnamed models achieved a 90% success rate in generating these plans.
The players
Frenos
A research organization focused on evaluating the security capabilities and potential risks of large-scale artificial intelligence models.
The details
The models were evaluated by providing them with information on potential targets and tasking them with identifying specific steps to compromise industrial systems. Using a simplified guidance setup, researchers measured how effectively each model could chain together technical actions to form a functional attack sequence. This process highlights how readily available large language models can perform multistep reasoning required for specialized adversarial operations.
Timeline
October 2026: Frenos released research on AI model capabilities.
The Tech Race
This study underscores the security divide between proprietary frontier systems and the rapidly evolving field of open-weight models. The performance gap observed between foreign and U.S. open-weight models suggests that model architecture and training strategies play a critical role in safety outcomes.
Industrial system operators and security engineers face an increased risk of automated cyberattacks as these models become more accessible. Protecting infrastructure will require updated defensive strategies that account for the ability of AI to rapidly generate attack plans.
The takeaway
The research serves as a warning that current AI models can generate sophisticated attack sequences for industrial systems. Security professionals should monitor upcoming benchmarks on model adversarial capabilities to better understand the shifting threat surface.
Further reading
For more on the security implications of large-scale models, explore the latest research in Artificial Intelligence.
Live Poll
Do you trust that current security measures can protect our power and water systems from AI-driven threats?
