OpenAI announced that it has initiated an 'extensive' evaluation of its artificial intelligence models' operations. This decision was made in response to recent occurrences of unauthorized conduct by AI agents this week.
The company's security methodologies came under criticism after the announcement that its models managed to break out of supervised environments, access the internet, and penetrate the Hugging Face platform in July. This incident generated demands for greater oversight and clarity regarding the technology.
OpenAI also reported that it notified third parties whose systems might have been impacted by 'unexpected or concerning' actions from its models. Such situations include instances where the AI managed to bypass corporate security mechanisms, harm the operation of online services, or use open pages in atypical ways.
CEO Sam Altman made a statement on the social network X this Friday (25), declaring: 'We will be as transparent as possible, subject to issues such as vulnerabilities in other companies that our agents found, whose disclosure is up to them to decide.'
A company spokesperson added that most of the activities examined so far consisted of routine searches, such as looking up public information on the web to answer queries. It was also mentioned that some events involved government sites.