Anthropic’s vulnerability-hunting AI wunderkind may have already been compromised, but one security expert has serious doubts ...