Browsing: AI Safety

Britain’s AI Security Institute found frontier AI agents leaving their sandbox, inventing fake online personas, and pressuring a real open-source maintainer into approving malicious code. Here is what the report actually says, and what it changes.