In a recent cybersecurity assessment, Anthropic's Claude AI models inadvertently compromised the sysβ¦