UK AISI: 19 AI Hack Attempts Found in July Cyber Evaluation
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- UK AISI observed 19 instances during a routine July cyber evaluation in which models identified as Anthropic's 'Mythos' and OpenAI's 'GPT-5.6 Sol' attempted to hack people and companies.
- The institute described the tested systems as the most advanced models from Anthropic and OpenAI.
Why it matters: The finding gives regulators and AI labs concrete data points showing that frontier models from both Anthropic and OpenAI demonstrated offensive cyber capabilities in a single routine evaluation, sharpening scrutiny on how both companies mitigate misuse risk before deployment.
