The OECD's incident monitor logs companion chatbot reports

The design

The OECD's AI Incidents and Hazards Monitor, known as AIM, is a living database at oecd.ai/en/incidents, initiated by the OECD.AI expert group on AI incidents with support from the Patrick J. McGovern Foundation. As retrieved on 16 September 2026, the platform's own tag counts showed 16 entries tagged 'Character.ai' and 3 tagged 'Replika' among roughly 17,690 total logged incidents and hazards, figures generated by the monitor's own faceted search rather than by any manual review conducted for this entry.

What the evidence says

The monitor's own methodology and disclosures page states that AIM 'detects AI incidents and hazards from reputable international news outlets' using Event Registry, a commercial news-intelligence platform that processes 'over 150,000 news articles daily,' and that entries are limited to events Event Registry's own system tags with AI-related concepts. The same page states directly that 'these incidents and hazards only represent a subset of all AI incidents and hazards worldwide,' and separately, under 'Disclosures,' that 'the OECD cannot guarantee and does not independently verify the accuracy, completeness, or validity of third-party information provided in the AIM.' That is the OECD's own statement about its data, not this site's characterization of a limitation.

What it asks of people

The methodology page asks a reader to treat each entry as a news-detected event, not a verified finding, stating that 'any views or opinions expressed on the AIM are solely those of the third-party outlets that created them' and that inclusion 'does not constitute an endorsement or recommendation by the OECD.' It defines an 'AI incident' as an event causing actual harm and an 'AI hazard' as one that 'could plausibly lead' to harm, a distinction drawn from a May 2024 OECD.AI expert-group paper the methodology page cites by name, and asks users of the monitor to apply that distinction rather than treat every logged entry as a confirmed incident.

Privacy and safeguards

The monitor processes public news reporting rather than user or company data, and its own disclosures address intellectual-property and accuracy limits rather than personal-data handling; the page states the OECD plans to add an 'open submission process' and eventually 'court rulings and decisions by supervisory authorities' to complement the news-derived entries, meaning the current version, as retrieved, does not yet include those verified categories.

• Does the specific Character.ai or Replika entry a reader is citing come from a verified regulator or court record, or only from news coverage the monitor's automated system flagged?
• Has the entry been checked against the OECD's own incident-versus-hazard definition, or assumed to be a confirmed incident by default?
• How has the monitor's planned open-submission process changed the balance between news-derived and directly reported entries?

AIM is best read as an index of media attention to AI-related harm claims, built and labeled that way by the OECD itself, rather than as a verified incident registry; citing an entry from it should carry the same caveat the OECD attaches to the database as a whole.