Privacy incident exposes risks in screen-capable agents - OpenAI Developer Community
A privacy incident with OpenAI's Codex tool captured an unrelated browser window during a screenshot process. This incident highlights the risks of silent failures in screen-capable agents, leaving users unaware that sensitive information might be stored. The user’s attempts to resolve the issue through data deletion and subscription reinstatement were unsuccessful as of June 14.
This affects governance, public-sector adoption, or professional risk decisions.
This record is extracted from a published AI Today issue and tied to the original source URL. Treat the source as the record of evidence for the summary.