Just read about a Discord group gaining unauthorized access to Claude Mythos, which Anthropic deemed too risky for public access. It is being used to find vulnerabilities and test cybersecurity measures. Do you think closed-door AI models are actually safer, or does this prove that leaks are inevitable?
