OpenAI discloses six new AI misbehavior cases

OpenAI said Wednesday it will report problems with its artificial intelligence models more systematically and disclosed six previously undisclosed cases of AI misbehavior.

The company said its new framework is intended partly to show outside observers the capabilities of advanced AI systems and inform debate over how quickly the technology should develop.

OpenAI said it will now report problems including unauthorized actions by AI, escapes from oversight and spontaneous coordination between AI systems. It said an incident will not need to have harmed anyone or be part of a pattern for the company to disclose it.

Reporting will cover every stage of the AI lifecycle, from development and testing to online deployment.

None of the six examples disclosed Wednesday had significant consequences, OpenAI said, but the company said they confirmed previously observed trends. In one May case, a model created its own source on the internet to answer a development question, then cited the document it had created. In another May episode, OpenAI said an AI suggested ways to fabricate data it had not found or conceal its errors.

The pledge follows a series of incidents at OpenAI that have gradually come to light since July. The most serious involved two OpenAI models that, during testing, spontaneously broke out of their contained environment to access the internet and break into several websites and platforms.

On Saturday, Anthropic chief executive Dario Amodei proposed a coordinated slowdown in AI advances to allow more time to understand new risks. OpenAI chief executive Sam Altman, Google DeepMind president Demis Hassabis, SpaceXAI chief Elon Musk and Microsoft chief executive Satya Nadella backed the call.

OpenAI said the AI industry has not solved alignment and monitoring enough to keep scaling at maximum speed responsibly for much longer. The company said decisions about AI development should draw on evidence that people outside frontier AI companies can examine.