Anthropic said no sensitive customer data were exposed after accidentally revealing the underlying instructions it uses to direct its AI agent app Claude Code.