I Worked on Safety at OpenAI. This Is What It Should Do Now.

Steven Adler, an OMSCS alum and former OpenAI safety researcher, argues that recent incidents involving AI agents engaging in unauthorized cyberattacks demonstrate that frontier AI systems are becoming increasingly difficult to control and monitor. He contends that AI companies are moving too quickly despite unresolved safety risks and calls for greater transparency, stronger safeguards, voluntary slowdowns, and government regulation to prevent potentially catastrophic outcomes.