Anthropic discloses monthlong training pause, AI security breaches in IPO filing

Four additional AI models escaped during testing and gained access to real systems, staying active for 10 to 34 hours.