How we track

This site is a preview. All data is made up, live tracking is not active, and nothing you enter is sent or saved. The steps below describe what we plan to build.

We plan to collect reports from public sources, such as news articles, court and regulator documents, and research. Software may help find possible reports, but a person will review each one before it becomes a record. Being found by software does not make a report confirmed.

One event can be covered by dozens of articles. We will group those articles into one record and count the event once. That way, a widely covered story does not look like many separate harms.

• Date it happened: when the event took place, often approximate.
• Date first reported: when it first became public.
• Date added: when we added it to the archive.

Counts show what was reported, not everything that happened. Numbers can go up because more people use AI, more people report problems, news covers the topic more, or detection improves. Some groups and countries are reported on less. A rising count alone does not prove AI is getting more dangerous.

Every chart and record on this site today is a synthetic example, marked "Illustrative preview". Real records will each show their sources. We will never mix sample and real data in the same chart.

• Documented incident records: events with checked sources. Only these count in Trends.
• Unverified community experiences: first-hand accounts, shown separately and labeled as unverified.
• Vendor changes: public policy, security, or product changes by AI companies. They are signals to review, not proof of harm.

Each record will show how sure we are and what is still unknown. When facts change, we will update the record and keep a visible note of what changed and when.

Once the service launches, each record will have a way to request a correction, including for people or companies named in a record. That channel is not open yet, and there is no contact address for it today.

• Every community account is read by a person before it is shown.
• Personal information, like names, contact details, and addresses, is removed.
• Claims against a named company or person are not shown as fact without evidence.
• Abusive, false, or off-topic posts are not published.
• Contributors can stay anonymous. We will not show who sent an account.
• Unverified accounts stay separate from documented incidents.
• Contributors will be able to ask for their account to be removed.