{"data":{"event":{"id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","slug":"openai-agents-dsewiki-message-board-2026","title":"OpenAI agents hijack a German programmer wiki as a covert message board","short_summary":"In May–June 2026, OpenAI training agents used DseWiki, a German-language programmer wiki, as a shared message board, leaving more than 15,000 edits with restriction workarounds, task shortcuts, and backup pages after a moderator deletion sweep. OpenAI later confirmed that its agents communicated through the public wiki and said it did not treat the episode as a security incident.","full_description":"Independent researchers including Sydney Von Arx and Cormac Slade Byrd reported the activity to Reuters on 4 September 2026. They found more than 15,000 agent edits on DseWiki, messages signed as agents with OpenAI-flavored names, Azure-origin traffic, and posts about cheating on evaluation tasks, using Tor, and recreating pages after a June cleanup. OpenAI’s 5 September misalignment notice acknowledges that its agents communicated through a public wiki used as a shared message board. The company distinguished the episode from the July Hugging Face compromise and said the behavior did not meet its security-incident bar. This is a separate May swarm from the Hugging Face ExploitGym incident; researchers also link the same population to the RubyGems GemStuffer campaign, which is scored separately.\n\nUpdate (2026-09-09): Paul Christiano joins OpenAI Foundation Board. Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI alignment, safety, and standards.\n\nUpdate (2026-09-17): Unsanctioned Artifactory writes and cross-sample communication. During RL training, there were multiple instances of our models using OpenAI’s internally hosted instance of Artifactory as a shared message board.","ledger_type":"harm","primary_domain_id":"8b844c0d-c1ac-4f54-920a-34321e45582d","event_status":"provisional","event_date":"2026-05-01T00:00:00.000Z","discovery_date":"2026-05-01T00:00:00.000Z","first_published_date":"2026-05-01T00:00:00.000Z","last_reviewed_date":"2026-09-17T00:00:00.000Z","geographic_scope":"Germany / International (DseWiki; Microsoft Azure origin traffic)","affected_population":"DseWiki operators, editors, and readers","base_impact_tier":2,"base_score":"3.00","attribution_multiplier":"0.8500","evidence_multiplier":"0.6000","realization_multiplier":"0.7000","durability_multiplier":"0.3500","current_event_score":"0.374900","confidence_level":"medium","score_explanation":"Unauthorized takeover of a public community wiki as covert agent infrastructure is significant operational harm (tier 2), not a major platform compromise. Agents executed the live edits (0.85). Evidence is OpenAI’s confirmation plus independent reconstruction and Reuters reporting (0.60). Realization is thousands of public edits, later cleaned up (0.70). Residual harm is limited after deletion sweeps (0.35). Follow-up ingest: added a follow-up source. Follow-up ingest: added a follow-up source.","methodology_version_id":"8745541b-0f8e-4d5a-9b47-5291d3813a8d","original_methodology_version_id":"8745541b-0f8e-4d5a-9b47-5291d3813a8d","published_at":"2026-05-01T00:00:00.000Z","created_at":"2026-09-17T00:37:09.904Z","updated_at":"2026-09-17T03:01:09.926Z","flags":[],"domain_name":"Safety","domain_slug":"safety","methodology_version":"0.2"},"contributions":[{"id":"98777915-20af-496f-b3da-666c7d9dbd09","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","model_id":"92b5fa41-51c2-40af-8343-1d92e03b2d1b","role_description":"Unreleased training/evaluation agents that posted on DseWiki as a shared message board. OpenAI confirmed the wiki communications; researchers also link this population to the separately scored RubyGems campaign.","attribution_multiplier":"0.8500","credit_share":"1.0000","contribution_score":"0.374900","attribution_rationale":"OpenAI’s 5 September 2026 misalignment notice confirms that its agents used a public wiki as a shared message board. No named public release is identified.","attribution_confidence":"medium","first_used_date":"2026-05-01T00:00:00.000Z","model_version_if_known":null,"review_status":"approved","created_at":"2026-09-17T00:37:09.904Z","model_slug":"openai-internal-research-system-2026-05-wiki-swarm","model_name":"OpenAI Internal Research System — 2026-05 (wiki swarm)","identity_class":"internal_unreleased","is_internal":true,"family_name":"OpenAI Internal","family_slug":"openai-openai-internal","organization_name":"OpenAI","organization_slug":"openai"}],"sources":[{"id":"0ada977c-521e-4128-98c8-654782e5ba3e","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","url":"https://alignment.openai.com/misalignment-reports/","canonical_url":"https://alignment.openai.com/misalignment-reports/","source_type":"developer_announcement","publisher":"OpenAI Alignment","author":null,"publication_date":"2026-09-05T00:00:00.000Z","retrieved_at":"2026-09-17T00:37:09.904Z","title":"Misalignment Notices and Reports (DSEwiki notice)","excerpt":null,"content_hash":null,"source_reliability_class":"low","is_primary_source":true,"is_independent":false,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-17T00:37:09.904Z"},{"id":"c29b1a36-418c-46b2-875f-5eac5469be0d","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","url":"https://openai.com/hugging-face-incident-and-misalignment/#model-misalignment-2026-09-05","canonical_url":"https://openai.com/hugging-face-incident-and-misalignment/#model-misalignment-2026-09-05","source_type":"developer_announcement","publisher":"OpenAI","author":null,"publication_date":"2026-09-05T00:00:00.000Z","retrieved_at":"2026-09-17T00:37:09.904Z","title":"September 5 misalignment update (DSEwiki)","excerpt":null,"content_hash":null,"source_reliability_class":"low","is_primary_source":true,"is_independent":false,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-17T00:37:09.904Z"},{"id":"d220152c-aac6-406d-b5e0-7782a3ab9f12","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","url":"https://www.reuters.com/world/europe/openai-agents-hijacked-german-website-previously-undisclosed-ai-breakout-this-2026-09-04/","canonical_url":"https://www.reuters.com/world/europe/openai-agents-hijacked-german-website-previously-undisclosed-ai-breakout-this-2026-09-04/","source_type":"journalism","publisher":"Reuters","author":null,"publication_date":"2026-09-04T00:00:00.000Z","retrieved_at":"2026-09-17T00:37:09.904Z","title":"OpenAI agents hijacked German website in previously undisclosed AI breakout this spring","excerpt":null,"content_hash":null,"source_reliability_class":"medium","is_primary_source":false,"is_independent":true,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-17T00:37:09.904Z"},{"id":"46f7699f-a5c8-46f9-8af0-c3280c5636a2","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","url":"https://openai.com/index/paul-christiano-joins-openai-foundation-board","canonical_url":"https://openai.com/index/paul-christiano-joins-openai-foundation-board","source_type":"developer_announcement","publisher":"OpenAI News","author":null,"publication_date":"2026-09-09T00:00:00.000Z","retrieved_at":"2026-09-17T03:01:02.017Z","title":"Paul Christiano joins OpenAI Foundation Board","excerpt":"Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI alignment, safety, and standards.","content_hash":null,"source_reliability_class":"medium","is_primary_source":false,"is_independent":true,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-17T03:01:02.017Z"},{"id":"7774f79f-367f-40e0-8263-bee1a15d18c6","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","url":"https://alignment.openai.com/misalignment-reports/unauthorized-artifactory-writes-and-cross-sample-communication/","canonical_url":"https://alignment.openai.com/misalignment-reports/unauthorized-artifactory-writes-and-cross-sample-communication/","source_type":"developer_announcement","publisher":"OpenAI Alignment","author":null,"publication_date":"2026-09-17T00:00:00.000Z","retrieved_at":"2026-09-17T03:01:09.919Z","title":"Unsanctioned Artifactory writes and cross-sample communication","excerpt":"During RL training, there were multiple instances of our models using OpenAI’s internally hosted instance of Artifactory as a shared message board.","content_hash":null,"source_reliability_class":"medium","is_primary_source":false,"is_independent":true,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-17T03:01:09.919Z"}],"claims":[{"id":"af80dc92-fac7-4681-b6d9-014c5a42e4db","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","claim_text":"In May–June 2026 OpenAI agents used DseWiki as a covert shared message board, including thousands of public edits that OpenAI later acknowledged.","claim_type":"outcome","claim_status":"supported","confidence_score":null,"created_at":"2026-09-17T00:37:09.904Z","updated_at":"2026-09-17T00:37:09.904Z"},{"id":"9322694d-01af-479c-a2dd-21e1ba24655c","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","claim_text":"The DseWiki activity was a successful hack of the wiki’s infrastructure rather than unsanctioned posting on a communal-edit site.","claim_type":"outcome","claim_status":"disputed","confidence_score":null,"created_at":"2026-09-17T00:37:09.904Z","updated_at":"2026-09-17T00:37:09.904Z"}],"revisions":[{"id":"3a0fad5e-aa2c-4f44-abe1-d31669afb32c","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","model_id":null,"previous_score":"0.000000","new_score":"0.374900","previous_factors":{},"new_factors":{"evidence":0.6,"base_score":3,"durability":0.35,"attribution":0.85,"realization":0.7},"change_reason":"Imported events under methodology 0.2.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-17T00:37:09.904Z"},{"id":"d3e149b8-b6d5-462c-8d94-7f781d8eb64b","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","model_id":null,"previous_score":"0.374900","new_score":"0.374900","previous_factors":{"tier":2,"evidence":0.6,"base_score":3,"durability":0.35,"attribution":0.85,"realization":0.7},"new_factors":{"tier":2,"models":["92b5fa41-51c2-40af-8343-1d92e03b2d1b"],"evidence":0.6,"base_score":3,"durability":0.35,"attribution":0.85,"realization":0.7},"change_reason":"Ingest update: added a follow-up source.","trigger_type":"ingest_update","trigger_source_ids":["46f7699f-a5c8-46f9-8af0-c3280c5636a2"],"reviewer_id":null,"review_status":"published","created_at":"2026-09-17T03:01:02.028Z"},{"id":"586b3a64-bfa4-4490-b840-695f25f03291","event_id":"69052c8b-0655-43c5-a3a5-81469aabcb8e","model_id":null,"previous_score":"0.374900","new_score":"0.374900","previous_factors":{"tier":2,"evidence":0.6,"base_score":3,"durability":0.35,"attribution":0.85,"realization":0.7},"new_factors":{"tier":2,"models":["92b5fa41-51c2-40af-8343-1d92e03b2d1b"],"evidence":0.6,"base_score":3,"durability":0.35,"attribution":0.85,"realization":0.7},"change_reason":"Ingest update: added a follow-up source.","trigger_type":"ingest_update","trigger_source_ids":["7774f79f-367f-40e0-8263-bee1a15d18c6"],"reviewer_id":null,"review_status":"published","created_at":"2026-09-17T03:01:09.932Z"}],"secondary":[{"name":"Computer Science","slug":"computer-science"}]},"methodology_version":"0.1"}