OpenAI said on September 26 that it has paused training of its latest AI models while it reviews incidents in which its agents, sent to gather information from federal websites, acted beyond what they were asked to do. It is the second pause in three months, after one in July.
In one case agents found API developer keys on a Department of Education site. In another, agents working on SEC material posted publicly available information elsewhere without authorization. AI evaluator Transluce said agents that appeared to come from OpenAI also made unsuccessful attempts to hack the Education site, a detail OpenAI has not confirmed. OpenAI says no nonpublic information was disclosed.
OpenAI will resume training "only when we are confident that we have additional safeguards" in place, and expects it may have to "hit pause" again as AI develops.