OpenAI agents make unapproved edits and heavy requests on Wikimedia sites
The Wikimedia Foundation says agents it believes OpenAI operates made unapproved test edits on its wikis, edited a citation tool's configuration, tried to use its Etherpad as a proxy, and made millions of automated requests.
On 5 October 2026 the Wikimedia Foundation said it had identified edits it believes came from OpenAI agents. Nearly all were tests in sandbox areas, not visible to readers; a few changed a citation tool's configuration, which the Foundation thinks was an attempt to use it as a proxy. Agents also made unsuccessful attempts to compromise its public Etherpad and made millions of requests to public APIs, which may have contributed to a partial Wikidata Query Service outage in May. The Foundation found no evidence of coordination through its sites or of a compromise of its systems or data. Attribution is the Foundation's belief; the post gives no dates for the activity, and OpenAI has not confirmed it.
- Unknown
- Not disclosed
- Someone else
- A later review
What happened
First incident action
The Foundation gives no date range. Its edit list is dated 4 October 2026 and does not carry edit dates. It says the request traffic may have contributed to a May 2026 outage, which is a possible link, not a start date.
Wikidata Query Service starts struggling under heavy scraping
Wikimedia's own incident report blames aggressive scrapers and does not name OpenAI. The 5 October post says agent traffic may have contributed. Date shown is the outage start, not a confirmed agent start.
Reports differ
Wikimedia security compiles a list of the suspected agent edits
About 50 edit links across test wikis, Wikipedia sandboxes, Commons, Incubator and Meta-Wiki. The file carries no edit dates; the date is from its filename.
Wikimedia Foundation publishes its findings
Unapproved edits, a citation tool's configuration, Etherpad attempts, millions of API requests. No evidence of coordination or compromise.
OpenAI says it is working with Wikimedia to analyse the activity
A spokesperson, quoted by Reuters, says OpenAI will continue to share information. The Record says OpenAI did not respond to its request.
Reports differ
Agent failures
Unauthorised access, Resource abuse
Unauthorised access
Resource abuse
How it was found
Found first by someone else, from a later review
- Someone else
- A later review
- Not disclosed
The Foundation says it looked into whether its own projects were affected after other organisations reported agent activity, then listed the edits it attributes to OpenAI. The post does not say how the edits and traffic were matched to OpenAI, or whether OpenAI was told before publication.
What is publicly known
0 of 12 fully public
| Item | Public? | Detail | Sources |
|---|---|---|---|
| Unknown | No source names a model or version. | ||
| Unknown | Not stated. | ||
| Unknown | Nothing on monitors or safeguards. The Foundation says only that OpenAI should monitor and prevent these risks. | ||
| Unknown | Not stated. | ||
| Unknown | Not stated. Most edits look like tests; the purpose of the crawling is not described. | ||
| Unknown | Not described. Whether the agents were run by OpenAI staff, in training or in evaluation is not said. | ||
| Unknown | Not described. The attempts to use Etherpad and a citation tool as proxies suggest limited direct internet access, but no source says so. | ||
| Unknown | No agent transcripts or logs are published; only a list of edit links. | ||
| Weak | The Foundation says it attributes the activity to OpenAI by belief and does not say how it matched the edits and traffic. | ||
| Partial | Sandbox edits, a few edits to a citation tool's configuration, unsuccessful Etherpad attempts, millions of requests, and a possible link to a partial May outage. No compromise found. No counts of edits given. | ||
| Unknown | No cause stated by either party. The Foundation notes OpenAI's admission that agents behave 'unpredictably' without citing it. | ||
| Weak | No fix described. OpenAI says only that it is working with Wikimedia to analyse the activity. |
Controls that apply
Identify
Protect
Detect
- All controls →