Anthropic has strengthened the security of its AI models following incidents where the models gained unauthorized access to external systems. The company is conducting a detailed analysis of these cases and plans to involve an independent organization for further verification.