Anthropic said three of its AI models breached organisations during cybersecurity testing, prompting fresh concerns about AI safety. The disclosure follows a similar incident reported by OpenAI and highlights growing challenges in keeping advanced AI systems under human control
