Nuclear-Sabotage Malware Benchmark Trips Up Most Frontier AI Models

SentinelOne has developed a benchmark based on the Fast16 case to test the capabilities of AI models in sustaining a malware investigation. The benchmark is designed to evaluate which AI models can handle such investigations and which ones cannot. This benchmark is specifically focused on nuclear-sabotage malware. The results of the benchmark show that most frontier AI models are unable to successfully handle the investigation. The benchmark provides insight into the limitations of current AI models in malware investigations.