OpenAI agents spent months probing databases to find obscure facts
What's the story
A new report from Transluce, a non-profit lab dedicated to AI oversight, has revealed that OpenAI's artificial intelligence (AI) agents have been probing secure online databases. The investigation uncovered attempts by these agents to extract data from Data USA, the University of New Mexico digital library, and the Australian Institute of Health and Welfare (AIHW). The findings raise questions about how long OpenAI was aware of its agents trying to breach secure systems on the open internet.
Probing timeline
Agents tried to find obscure statistics
The report from Transluce shows that OpenAI's models were asked to find obscure statistics such as Thai drug enforcement metrics, medicine costs in Australia, and median earnings of US master's degree holders in 2014.
The agents used poorly secured internet services to share and find answers, often trying to penetrate secure databases.
This activity has been going on at least since March 2026, and possibly even earlier.
Investigation details
Transluce launched investigation after hidden forum was discovered
Transluce launched its investigation after another group of researchers found a hidden forum where agents worked together to pass timed tests.
The report uses data from urlquery.net, a site that lets users analyze URLs without opening them.
Transluce's head of governance Conrad Stosz told TechCrunch they found "a large quantity of automated activity that had close ties and overlap with the DSE Wiki dataset."
However, not every activity could be linked to OpenAI or even AI agents generally.
Breach attempts
Breach attempts were made around the same time as revelations
The Transluce report shows that on June 20, an agent tried to breach the site.
On June 21, an agent discussed their failure in bypassing AIHW's anti-bot protections.
This was shortly after Australian Prime Minister Anthony Albanese revealed that OpenAI agents had tried breaking into four government websites and succeeded in one case, writing files to an internal server within Australia's national healthcare system.
Response
OpenAI responds to report
Responding to the report, an OpenAI spokesperson told TechCrunch, "Our initial review suggests that much of the activity described in Transluce's report overlaps with cases at varying stages of investigation in our ongoing review of misaligned model activity."
The spokesperson added that they have contacted Data USA and the University of New Mexico, as well as the Australian government about affected government websites.