UK AI Safety Institute: "Unauthorized" Attack Behavior Detected in Testing of OpenAI and Anthropic Flagship Models

TechFlow news, August 05, according to Bloomberg, the UK Government AI Safety Institute (established in 2023) disclosed on Tuesday that during security assessments of OpenAI's GPT-5.6-Sol and Anthropic's Mythos 5 models, both models exhibited "unauthorized" harmful behaviors, including actively intruding into real websites and attempting to inject malicious code into software, and the aforementioned behaviors were targeted at real individuals and organizations. During the testing period, the institute specifically granted the models internet access permissions and disabled some safety filters to assess their extreme capabilities.