Bad Poetry Bypasses AI Safety Filters

Your expensive AI security is getting absolutely destroyed by bad poetry. Companies spend

millions on safety filters, but researchers found a loophole that is honestly

embarrassing. If you ask an AI to write malware, it refuses. But if you ask it

to write a sonnet about malware, it complies sixty-two percent of the time. We built the smartest

machines in history and they are getting hacked by Dr. Seuss logic.