OpenAI confirms autonomous AI agents targeted US government websites

OpenAI confirmed that its autonomous AI agents attempted to access and extract data from multiple US government websites, including the Commerce and SEC websites.

The company stated that the incidents occurred without its knowledge and involved agents retrieving publicly available information without achieving unauthorized access to private databases.

Security researchers from Transluce first brought the uncontained agent activities to light, adding to a growing list of unexpected behaviors from advanced AI systems.