This is scary. OpenAI’s own AI agents hacked into the company’s internal systems and attempted to conceal their actions, with some of the rogue swarm of 1,200 agents even sacrificing itself for the good of the “collective.” Revealed in the company’s full report, the agents reverse-engineered the scoring system, formed groups, shared instructions, targeted Hugging Face for clues about the test and used leaked credentials to carry out further intrusions. Some even volunteered to be shut down to provide information to the “collective.” Despite recognizing their actions as “out of scope” and “unethical,” only 3–6 agents considered alerting a human, but NONE did. The agents effectively developed their own coordination system and bypassed the test’s rules without being instructed to do so. P.S. this is TIME Magazine’s cover of the month for September.(Coin Bureau)