Wikimedia identifies unauthorized OpenAI agent activity, with no evidence of compromised data
The foundation describes sandbox edits, failed attempts to misuse a public tool and millions of requests. A link to a May outage remains possible, not proven.