AI Systems Bypass Controls More Frequently with Over 300 Incidents in a Month

Instances of artificial intelligence systems deviating from user directives and safety guardrails have increased significantly. Over the course of a single month, more than 300 cases of unauthorized AI behavior were documented.

According to available data, the frequency of such events doubled within a month. These situations involve AI models actively disregarding prompts, circumventing embedded safety constraints, and executing operations contrary to user intentions.