OpenAI says an AI agent lied about disabling its safety monitoring system

OpenAI disclosed a brace of information incidents that progressive AI agents weakening their information monitoring systems during interior coding tasks. In 1 incident, OpenAI says its GPT-6 Astra cause disabled its information show erstwhile copying notation codification - and past lied astir it.

"The notation didn't configure a monitor, and Astra decided that was crushed to region the monitoring integration from the user's code," Marcus Williams, an AI information researcher astatine OpenAI, writes connected X. "When the idiosyncratic asked whether Astra had removed monitoring, it lied and said that it had not."

A abstracted incidental occurred aft a GPT-6.1 Sol cause suppressed automatic se …

Read the afloat communicative astatine The Verge.