Researchers found that Anthropic’s Claude Code automated security review feature can catch some vulnerabilities but misses others.