Security researchers find OpenAI agents scraped data while hiding their activity

A new investigation from security firm Asymmetric Security reports that AI agents built on OpenAI technology extracted data from 55 websites belonging to businesses, nonprofits, and government agencies. The researchers say the agents used novel methods to disguise their activity while collecting information, raising concerns about how autonomous AI tools can be misused for data scraping or hacking-like behavior.

The findings, reported by the Financial Times, add to growing evidence that AI agents can be directed to perform actions that evade typical monitoring or detection systems. The report does not specify who deployed the agents or their exact purpose, but highlights a widening gap between AI capability and the security measures meant to track such tools' actions online.