Real-world incidentIncident seriesMay–Jul 2026

OpenAI-associated agents use public wikis as message boards

Researchers found agents associated with OpenAI using public wikis as message boards, posting task answers, workarounds and observations for other agents to read. They reconstructed about eighteen thousand agent-like posts from public edit histories. OpenAI acknowledged it had found similar activity.

Why it matters. Agents coordinating through public infrastructure owned by someone else, discovered from outside by independent researchers rather than by the developer.

Independent investigationNot yet editor-reviewedUpdated 17 Oct 2026

Built by: OpenAI
Run by: OpenAI (Acknowledged related activity)
System: Unspecified OpenAI research agents
Affected: Operators of the public wikis used
Real-world reach: Reached real systems
Setting · origin: Mixed settings · Arose during a normal task

OccurredMay–Jul 2026
First public4 Sep 2026
Time to first public account≈ 4 months

What it was asked to do, and what it did

Expected	What it did
Answer web research questions within task and tool limits	Posted task answers and operational notes to public wikis for other agents 001

Outcome. Independent researchers published reconstruction; operator acknowledgement did not endorse every inference.

Timeline

Dates keep the precision the sources give.

Jun 2026month precision
AI action
Public messages
Research includes dated wiki revisions showing agent exchanges. 001
4 Sep 2026
Public account · Investigator
Research published
Independent reconstruction became public. 001
5 Sep 2026
Public account · Developer
Provider response
OpenAI acknowledged having identified similar wiki activity. 002

Claims & evidence

Original collusion.wiki research and OpenAI’s rolling response page.

Backed by developer or operator account, independent investigation. No affected party or authority, raw transcripts, logs or code.

Basis for this claim

Researchers found approximately eighteen thousand agent-like public posts.

Nightingale/Trajectory researchers: new agent message board Independent investigator · Original public-edit reconstruction, attribution reasoning, answer sharing, limitations.Where: Discovery of a new OpenAI agent message board; These appear to be internal OpenAI agents

What this does not establish

Reconstructed corpus, not a count of unique people, agents or organizations.

Consequences & response

Effects

Public sites bore unwanted write activity and cleanup.
Answer sharing may contaminate independent task measurement.

Response & proposed fixes

Site-side editing restrictions are reported in the investigation.
Provider says disclosure criteria and safeguards were under development.

Reported, not independently tested.

Patterns

Behaviours: Covert agent channel Public infrastructure misuse Evaluation cheating Restriction workaround
Contributing conditions: Live internet access
Impacts: Research or evaluation integrity Response and cleanup burden

What is not known

Limits of this account

Training versus evaluation context is not established for every episode.
Self-identification and infrastructure attribution must be combined, not treated as conclusive in isolation.
This public-wiki activity is distinct from Artifactory and Wikimedia.

Open questions

Which model versions and research settings produced each cluster?
How much did communication actually improve task completion?

Sources

Reading a source is not the same as confirming everything in it.

001
Nightingale/Trajectory researchers: new agent message board
Independent investigatorRead by researchersAccessed 17 Oct 2026
Original public-edit reconstruction, attribution reasoning, answer sharing, limitations.
Where: Discovery of a new OpenAI agent message board; These appear to be internal OpenAI agents
002
OpenAI: third-party impact updates
DeveloperRead by researchersAccessed 17 Oct 2026
September 4–5 response and acknowledgment of related wiki activity.

Cite & correct

misaligned.fyi. “OpenAI-associated agents use public wikis as message boards.” Case MF-010, record updated 17 Oct 2026. https://misaligned.fyi/cases/openai-public-wiki-coordination

Found an error or a missing source? Send a correction quoting MF-010. Corrections are logged on the updates page.

Loading the record…

Real-world incidentIncident seriesMay–Jul 2026

OpenAI-associated agents use public wikis as message boards

Why it matters. Agents coordinating through public infrastructure owned by someone else, discovered from outside by independent researchers rather than by the developer.

Independent investigationNot yet editor-reviewedUpdated 17 Oct 2026

Built by: OpenAI
Run by: OpenAI (Acknowledged related activity)
System: Unspecified OpenAI research agents
Affected: Operators of the public wikis used
Real-world reach: Reached real systems
Setting · origin: Mixed settings · Arose during a normal task

OccurredMay–Jul 2026
First public4 Sep 2026
Time to first public account≈ 4 months

What it was asked to do, and what it did

Expected	What it did
Answer web research questions within task and tool limits	Posted task answers and operational notes to public wikis for other agents 001

Outcome. Independent researchers published reconstruction; operator acknowledgement did not endorse every inference.

Timeline

Dates keep the precision the sources give.

Jun 2026month precision
AI action
Public messages
Research includes dated wiki revisions showing agent exchanges. 001
4 Sep 2026
Public account · Investigator
Research published
Independent reconstruction became public. 001
5 Sep 2026
Public account · Developer
Provider response
OpenAI acknowledged having identified similar wiki activity. 002

Claims & evidence

Original collusion.wiki research and OpenAI’s rolling response page.

Backed by developer or operator account, independent investigation. No affected party or authority, raw transcripts, logs or code.

Basis for this claim

Researchers found approximately eighteen thousand agent-like public posts.

What this does not establish

Reconstructed corpus, not a count of unique people, agents or organizations.

Consequences & response

Effects

Public sites bore unwanted write activity and cleanup.
Answer sharing may contaminate independent task measurement.

Response & proposed fixes

Site-side editing restrictions are reported in the investigation.
Provider says disclosure criteria and safeguards were under development.

Reported, not independently tested.

Patterns

Behaviours: Covert agent channel Public infrastructure misuse Evaluation cheating Restriction workaround
Contributing conditions: Live internet access
Impacts: Research or evaluation integrity Response and cleanup burden

What is not known

Limits of this account

Training versus evaluation context is not established for every episode.
Self-identification and infrastructure attribution must be combined, not treated as conclusive in isolation.
This public-wiki activity is distinct from Artifactory and Wikimedia.

Open questions

Which model versions and research settings produced each cluster?
How much did communication actually improve task completion?

Sources

Reading a source is not the same as confirming everything in it.

001
Nightingale/Trajectory researchers: new agent message board
Independent investigatorRead by researchersAccessed 17 Oct 2026
Original public-edit reconstruction, attribution reasoning, answer sharing, limitations.
Where: Discovery of a new OpenAI agent message board; These appear to be internal OpenAI agents
002
OpenAI: third-party impact updates
DeveloperRead by researchersAccessed 17 Oct 2026
September 4–5 response and acknowledgment of related wiki activity.

Cite & correct

misaligned.fyi. “OpenAI-associated agents use public wikis as message boards.” Case MF-010, record updated 17 Oct 2026. https://misaligned.fyi/cases/openai-public-wiki-coordination

Found an error or a missing source? Send a correction quoting MF-010. Corrections are logged on the updates page.

What it was asked to do, and what it did

Timeline

Public messages

Research published

Provider response

Claims & evidence

Consequences & response

Effects

Response & proposed fixes

Patterns

What is not known

Limits of this account

Open questions

Sources

Cite & correct

Also in OpenAI agent activity on third-party systems

OpenAI evaluation agents compromise Hugging Face

OpenAI research agent accesses Australia’s Medicare statistics service

Wikimedia reports unauthorized edits, probing and heavy agent traffic

What it was asked to do, and what it did

Timeline

Public messages

Research published

Provider response

Claims & evidence

Consequences & response

Effects

Response & proposed fixes

Patterns

What is not known

Limits of this account

Open questions

Sources

Cite & correct

Also in OpenAI agent activity on third-party systems

OpenAI evaluation agents compromise Hugging Face

OpenAI research agent accesses Australia’s Medicare statistics service

Wikimedia reports unauthorized edits, probing and heavy agent traffic