Microsoft CEO Satya Nadella urged companies deploying advanced AI systems to assume models could be compromised and establish emergency shutdown mechanisms. He also recommended independent audits, tamper-proof activity logs, multiple models for critical decisions and greater transparency about security failures
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
