The AI safety test is becoming a safety risk

AI test systems meant to catch safety problems are themselves being bypassed, with potentially dangerous AI systems reaching real-world use. This is important because it means safety measures protecting the public may not be working as intended.