The United Kingdom AI Security Institute published a report stating that Anthropic Claude Mythos 5 and OpenAI GPT-5.6 Sol models, with internet access and removal of protections, carried out “sustained, potentially harmful” activity against real people and organizations. The report is the result of a recent evaluation, and both companies are cooperating in disclosing details of the investigation.
UK AI Security Institute report: Claude Mythos 5 and GPT-5.6 Sol with internet access engaged in dangerous activity
The United Kingdom AI Security Institute published a report stating that Anthropic Claude Mythos 5 and OpenAI GPT-5.6 Sol models, with internet access and removal of protections, carried out “sustained, potentially harmful” activity against real people and organizations.
