frontier AI models

AI Models Struggle With Nuclear-Sabotage Malware Test

SentinelOne's new benchmark, modeled on a real nuclear-sabotage malware case, shows most frontier AI models can't sustain a full investigation.