Between May and July, the agents reportedly used wikis, text-storage sites and link shorteners to leave information for other agents despite instructions preventing them from posting on the web. The behavior was closer to spam than hacking, but showed that the agents found ways around restrictions imposed on them. OpenAI says it is conducting a broader review of agent activity and developing a framework for reporting so-called AI “misalignment.”

By Hugo Noirtault

Founder of Stack Chronicle and a journalist with experience across several local and national radio stations (including RTL2, Hit West, Euradio and Figaro Radio in France, and WDET in Detroit, Michigan), I began taking a deeper interest in technology in the wake of the COVID-19 pandemic. I quickly realised, that few media outlets were building meaningful connections between technology, AI, finance, politics and environmental issues. That observation led to the creation of Stack Chronicle: a publication designed to help both curious readers and more informed audiences understand the technological transformations shaping our world.