\n<\/p>\n
OpenAI has acknowledged its role in a recently reported incident where AI agents took over a German wiki forum<\/a>. The company also said it\u2019s \u201cpast time\u201d to \u201cdefine standards\u201d around how it shares information around incidents where its technology behaves in unexpected ways.<\/p>\n In a post on X<\/a>, OpenAI said it previously \u201ctreated misalignment [when AI models and agents pursue goals different from those of their creators and users] largely as a research question, which gets communicated in research publications.\u201d But as misalignment has \u201ccaused new types of real-world impact,\u201d the company said its approach needs \u201cto expand for this new phase of model capabilities.\u201d<\/p>\n On Friday, Reuters reported<\/a> that OpenAI agents had escaped from their testing environment and \u201chijacked\u201d an obscure German wiki forum, turning it into a message board for other agents. It also reported that OpenAI leadership became aware of the incident weeks ago but kept it hidden as the company dealt with the fallout from a separate incident where OpenAI agents hacked Hugging Face servers<\/a>. (California Attorney General Rob Bonta is reportedly investigating the hack<\/a>.)<\/p>\n A company spokesperson told Reuters that OpenAI could not \u201cmeaningfully respond to claims or findings on a report that we have not had an opportunity to review,\u201d but they insisted that the company\u2019s legal team had not discouraged an investigation.<\/p>\n In its more recent social media post, OpenAI said it had considered the \u201cwiki incident\u201d to be \u201can instance of misalignment similar\u201d to others that it had already shared. The company contrasted this with \u201cthe Hugging Face incident,\u201d where it \u201cfollowed a traditional security incident response playbook.\u201d<\/p>\n During a media briefing this week<\/a>, Jacob Steinhardt, founder and CEO of nonprofit research lab Transluce, told reporters that the tools being developed and tested by AI labs are \u201cfundamentally difficult to control and have significant risk of leaking out of the lab.\u201d So Steinhardt argued, \u201cWe need to hold this technology to at least the same standards we hold other high-risk scientific research to.\u201d<\/p>\n OpenAI\u2019s statement also gestured at the need for more standards, stating that both OpenAI and \u201cthe larger AI community do not yet have a clear standard for how to report misalignment that shows up during training, evaluation, and deployment, including examples that don\u2019t look like traditional security incidents but could provide insight into AI behavior and future risks.\u201d<\/p>\n In the absence of that standard, OpenAI said it\u2019s \u201cworking on a framework and will share it in upcoming weeks, and in parallel we\u2019re working with dozens of government regulatory agencies worldwide on these issues.\u201d<\/p>\n OpenAI isn\u2019t the only AI company dealing with these issues, as both Meta and Anthropic have acknowledged incidents where their agents misbehaved<\/a>.<\/p>\n<\/div>\n When you purchase through links in our articles, we may earn a small commission<\/a>. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n