Testers Got Around OpenAI's AI Watchdog in Three Weeks

Outside testers found six ways to get around a safety monitoring system used by OpenAI within three weeks. The headline suggests the fine print of such safeguards matters as much as the system itself.