An example from the tracker
Promised at the AI Seoul Summit, May 2024, to publish a safety framework before the Paris summit on 10 February 2025.
- Status
- Delivered on time
See who keeps their word.
We research what companies and public figures have promised, predicted and claimed, and score how often their word has held up. Every verdict links to its evidence.
Track what management promised, and what it delivered.
- Equity. Has this management team delivered the launches, openings and targets it announced?
- Credit. Did the company deliver past deleveraging, disposal and spending promises?
- ESG. Are its climate and sustainability targets being met, with evidence?
Free for research in the public interest.
- Track commitments. Did AI labs publish the safety frameworks they promised? Are companies meeting their climate pledges?
- Check before you fund. The public record of an applicant, and of the organisations and people its application names.
- Check before you publish. A public figure’s or organisation’s record of promises and predictions, with sources you can cite.
Request free access
Public organisations and public figures only. Scope
An example from the tracker
Ørsted
“We have decided to stop using coal altogether at our power plants from 2023.”
- Status
- Resolved false
- Why
- Danish authorities ordered three units kept running during the energy crisis; the exit moved to 2025.
- Evidence
- Ørsted interim reports, 2022 and 2023
Also on its record
Read the Ørsted case study
- Sell half of Hornsea 1
- Largest offshore wind farm
- Certified biomass
How it works
Find what they said.
Promises, predictions and factual claims from their public professional output.
Check each one against the record.
The test is fixed before any research. Two independent model checks; disagreements are flagged.
Show the evidence.
Every verdict comes with the quote, the sources and the reasoning. Where the record is too thin, it says so instead of guessing.
Every claim gets one of six answers
The more of the circle is filled, the more of the claim held up.
- Resolved trueIt happened as said. For a promise: delivered on time.
- Mostly trueIt happened, with a gap the evidence records.
- MixedParts held up and parts did not.
- Mostly falseSomething related happened, not what was said.
- Resolved falseIt did not happen as said, or not by the date.
- UnresolvableThe record is too thin to decide, so there is no guess.
Case studies
-
Google IBM MicrosoftPublic-interest research
Did AI companies publish the safety frameworks they promised?
AI companies made the same promise at the Seoul summit. We checked each one, then compared with an independent list.
-
2017 2023Financial research
Did Ørsted do what it said?
Coal exit, wind farms, earnings forecasts: a company’s commitments over a decade, each with its outcome.
-
2013 2026Public-interest research
Thirteen years of self-driving predictions
Elon Musk’s dated predictions about self-driving Teslas, checked one by one against what happened.
All case studiesHow they were tested against published outcomes
Why trust it
- A published method. Anyone can read how every verdict is reached.
- Evidence for every verdict. Quotes are matched word for word to their source.
- “Unresolvable” rather than a guess. A thin record is reported as thin.
- A right of reply. Anyone with a report can answer any claim, and the reply is shown on it. If a report is about you
Scope
Public organisations and public figures only, from their public professional work. No politicians in office and no party-political claims. Machine verdicts are labelled until a person has reviewed them.