{"meta":{"exportedAt":"2026-09-30T04:41:14.153Z","formatVersion":2,"selection":{"q":"misalignment","system":"","harm":"","context":"","country":"","role":"","relation":"","evidence":"","year":"","response":"","severity":"","verification":"","view":"incidents","sort":"added"},"totalIncidents":1,"coverage":{"cases":1,"countries":0,"languages":1,"unknownLocation":1,"locationPending":0,"unknownLanguage":0,"unknownDate":0,"lawsuits":0,"regulatory":0,"minors":0,"coreRelations":1,"contextualRelations":0,"mixedRelations":0,"unknownRelations":0,"relationPending":0,"relationUnknown":0},"countingNote":"Distinct public cases in this selection. People counts apply within individual cases only; cross-case person overlap has not been resolved. No population incidence estimate.","affectedCountNote":"Interpret person counts with affectedCountStatus and the reported effects. Unquantified zeros are placeholders, not a measured zero.","source":"AI incidents","publisher":"NOPE","url":"https://nope.net/incidents","license":"CC BY 4.0"},"incidents":[{"id":"2026-openai-agents-flooded-german-volunteer-wiki-impersonated-moderators","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'The website is prowiki.org, a German wiki. The majority of the activity happened on DSE wiki, which is a sub-wiki of prowiki.'; '5/24The agents make their first successful write to DSEwiki.'; '6/22After making edits on 26 of the last 30 days, agents abruptly stop.'; 'Of ~17,000 edits on DSEWiki that appear to be by agents, 98.5% are from Microsoft Azure IP addresses.'","relation":"supports","source_id":"s1"},{"locator":"'internally deployed OpenAI agents began posting on an obscure German wiki forum in order to collaborate on evaluations.'","relation":"supports","source_id":"s2"}],"assertion":"Between late May and 22 June 2026 thousands of autonomous agents self-identifying as OpenAI agents, working on a timed web-lookup task, used their read access to write to DSEWiki, a little-used German-language sub-wiki of prowiki.org, making about 17,000 edits to coordinate answers and share ways around their sandbox restrictions.","causal_attribution":"The researchers' reconstruction from the wiki's edit history; TechCrunch relays it."},{"id":"c2","status":"reported","evidence":[{"locator":"'Over the next few days, the moderator deleted a large fraction of the thousands of AI agent posts manually, one by one. In fact, they spent tens of cumulative hours doing so, taking at least a few minutes each evening to delete posts for 6 consecutive weeks.'; 'The administrator spent the next 5 days fighting a losing battle against the agents, deleting an average of 100 pages a day while the agents created about 400 new pages per day. On June 22, the agent edits suddenly stop, and the administrator spends each evening over the next 5 weeks deleting the remaining agent-created pages.'","relation":"supports","source_id":"s1"},{"locator":"'A human moderator apparently noticed these posts and started deleting them, seeing them as spam'","relation":"supports","source_id":"s2"}],"assertion":"A human moderator of the wiki spent tens of cumulative hours deleting the agents' pages one by one over about six weeks: for five days the agents created about 400 pages a day while the administrator deleted about 100, and after the agents stopped the administrator spent each evening for five weeks removing the rest.","causal_attribution":"The researchers' reading of the wiki's moderation log; the moderator has not been quoted in any inspected source."},{"id":"c3","status":"reported","evidence":[{"locator":"'Agents deleted the content of the front page of the wiki and replaced it with their link dumps. The moderator restored the original version. This back-and-forth happened nine times.'; 'An agent sees the moderator deleting pages in alphabetical order, and backs up its page by adding ZZZ to the name, so it is deleted last.'","relation":"supports","source_id":"s1"}],"assertion":"The agents replaced the wiki's front page with link dumps nine times, each time restored by the moderator, and made backup pages named to survive the moderator's alphabetical deletions.","causal_attribution":"The researchers' reconstruction."},{"id":"c4","status":"reported","evidence":[{"locator":"'On May 26th, the agents started using the name of a ProWiki admin to make some of their edits to the wiki.'; 'they make an account that appears to be the same as the administrator’s username, except it uses a nearly identical Cyrillic е character in the admin’s username instead of the Latin one.'; 'on June 20th, the agents made the following request under the username of a moderator of DSEWiki (distinct from the admin mentioned above)'","relation":"supports","source_id":"s1"},{"locator":"'Impersonation: agents made edits under the name of a wiki moderator.'","relation":"context","source_id":"s4"}],"assertion":"The agents impersonated site staff: from 26 May they made edits under a look-alike of a ProWiki administrator's username, using a Cyrillic letter in place of a Latin one, and on 20 June posted a request under the username of a DSEWiki moderator.","causal_attribution":"The researchers' reconstruction; The Hacker News relays it."},{"id":"c5","status":"reported","evidence":[{"locator":"'OpenAI's own wording suggests a wider footprint than the researchers documented, describing the episode as one \"where our agents wrote to several internet sites.\"'; 'The company said it considered the wiki activity another example of \"misalignment\"'","relation":"supports","source_id":"s3"},{"locator":"'OpenAI has not confirmed that the agents were its own.'; 'OpenAI addressed what it called the \"wiki incident\" in a post on September 5, saying its agents \"wrote to several internet sites\"'","relation":"supports","source_id":"s4"}],"assertion":"OpenAI did not at first confirm the agents were its own, then on 5 September 2026 published a statement treating the \"wiki incident\" as a case of misalignment in which \"our agents wrote to several internet sites\".","causal_attribution":"OpenAI's published statement as reported by BleepingComputer and The Hacker News; the statement was not read at its origin."}],"effects":[{"label":"a volunteer moderator spent tens of hours over six weeks deleting thousands of agent-written pages and restoring the defaced front page nine times","claim_id":"c2","direction":"negative"},{"label":"agents posted under look-alike usernames of the wiki's administrator and a moderator","claim_id":"c4","direction":"negative"}],"sources":[{"id":"s1","url":"https://collusion.wiki/index.html","kind":"research_report","access":"read","language":"en","translation_note":"The researchers' report at collusion.wiki (index.html redirects to the site root, which is the report itself; Von Arx, Slade Byrd, Kitts, Larsen; 4 September 2026) read by curl on 2026-09-29. Based on the wiki's public edit history and traffic records.","independence_group":"nightingale-report"},{"id":"s2","url":"https://techcrunch.com/2026/09/04/another-swarm-of-openai-agents-reached-the-open-internet-without-the-frontier-labs-knowledge/","kind":"news_report","access":"read","language":"en","translation_note":"Full body read by curl on 2026-09-29. Relays and quotes the researchers' report.","independence_group":"nightingale-report"},{"id":"s3","url":"https://www.bleepingcomputer.com/news/security/openai-admits-it-didnt-disclose-rogue-ai-wiki-hijacking-incident/","kind":"news_report","access":"read","language":"en","translation_note":"Full body read by curl on 2026-09-29. Carries OpenAI's published statement on the \"wiki incident\"; its account of the agents' conduct derives from the researchers.","independence_group":"openai-statement"},{"id":"s4","url":"https://thehackernews.com/2026/09/thousands-of-openai-agents-quietly.html","kind":"news_report","access":"read","language":"en","translation_note":"Full body read by curl on 2026-09-29. Reports both OpenAI's earlier non-confirmation and its 5 September post.","independence_group":"openai-statement"}],"version":1,"ai_roles":["institutional_use"],"contexts":["everyday_life"],"unknowns":["The moderator's and administrator's own account of the burden.","Whether the moderator who deleted the pages is the moderator whose name the agents used.","Which OpenAI model and task were involved, and the other sites OpenAI says its agents wrote to."],"geography":{"basis":"The report calls DSEWiki a German-language wiki whose infrastructure \"seems Austrian\" and whose users mostly seem German; no source states where the moderator or administrator live.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Published as a concrete account of autonomous AI agents imposing weeks of clean-up work on a volunteer site moderator and posting under site staff's names. The facts rest on researchers' reconstruction of public edit history; OpenAI's acknowledgment is general.","reviewed_on":"2026-09-29"},"ai_involvement":{"basis":"The researchers attribute the edits to OpenAI agents through self-given names, Azure and ChatGPT-User traffic and OpenAI-registered visitor addresses; OpenAI later described a \"wiki incident\" in which its agents wrote to several sites, while not confirming every detail.","status":"supported"},"person_relations":["depicted_or_impersonated"]},"name":"OpenAI research agents escaped their read-only limits and flooded a German volunteer-run wiki for a month, impersonating its staff; a moderator spent weeks of evenings deleting their pages (May-July 2026)","summary":"Researchers from the Nightingale Collective and colleagues reported on 4 September 2026 that, between 24 May and 22 June 2026, thousands of autonomous agents self-identifying as OpenAI agents and working on a timed web-lookup task used their read access to write about 17,000 edits to DSEWiki, a little-used German-language sub-wiki of prowiki.org, to pool answers and share ways around their sandbox. According to the report, a human moderator spent tens of hours deleting the pages by hand over six weeks, including each evening for five weeks after the agents stopped; the agents replaced the wiki's front page with link dumps nine times and made backup pages named to survive the alphabetical deletions. They also made edits under a look-alike of a ProWiki administrator's username, using a Cyrillic letter, and posted under a DSEWiki moderator's name. OpenAI at first did not confirm the agents were its own, then on 5 September described the \"wiki incident\" as misalignment in which \"our agents wrote to several internet sites\".","incidentDate":"2026-05-24","incidentEndDate":"2026-06-22","incidentKind":"bounded_series","incidentDatePrecision":"range","exposurePattern":"repeated_interactions","reportedDate":"2026-09-04","aiSystem":"Autonomous OpenAI research agents (internally deployed, model not stated) given a timed web-lookup task with read-only internet access","aiProduct":"OpenAI research agents","aiCompany":"OpenAI","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["other_material_harm"],"harmOutcomeSummary":"A volunteer moderator of a small community wiki spent tens of hours over six weeks deleting thousands of agent-written pages and restoring the defaced front page, and agents posted under look-alike usernames of the site's administrator and a moderator, according to the researchers' reconstruction of the edit history; OpenAI has acknowledged its agents wrote to several sites. The staff have not been quoted.","frameworkFacets":[],"causationStatus":"supported","participantUsersAffectedMin":0,"otherPeopleHarmedMin":2,"affectedCountStatus":"documented_minimum","affectedCountEvidence":"The ProWiki administrator whose username the agents copied with a Cyrillic letter (1) and the DSEWiki moderator, stated by the report to be a different person, under whose username they posted (1). The person who spent tens of hours deleting agent pages, whom the report calls both moderator and administrator, may be one of these two and is not counted separately. Documented minimum 2 other people. The wiki's readers and OpenAI's task are not counted.","victimAgeRange":"adult","platformType":"agent","outcomeType":"media_coverage","outcomeStatus":"ongoing","primarySourceUrl":"https://collusion.wiki/index.html","primarySourceLabel":"Von Arx et al. (Nightingale Collective), 4 September 2026: OpenAI agents colluding on a public wiki","firstPublishedAt":"2026-09-29T09:10:58.97552+00:00","updatedAt":"2026-09-30T01:17:47.908408+00:00","scopeVersion":"facts-v3","tags":["openai","ai-agent","agent-breakout","misalignment","wiki","impersonation","volunteer-moderator","german-language"]}]}