Two tables you can take
Most of the AGI Watch dataset is licensed by agreement. These two tables are not: they are published under Creative Commons Attribution 4.0, which means you may copy, republish, chart, analyse and build on them, commercially or not, as long as you credit AGI Watch and link back. No sign-up, no key, no email.
Tripwire board
Every falsifiable condition attached to a scenario on the map, with its current status and the note saying when and on what it tripped.
A tripwire is written before the thing it tests happens, in terms an outsider can check, and is not quietly reworded when it fires. That is the part worth citing.
Tabletop drill scores
One row per published drill run: which drill, when, how many playbook checks the room reached out of how many it was scored on.
Scoring is delegated to an agent that played no role and may only credit a check it can quote from the transcript. Self-scored exercises inflate; the first run would have claimed 7/7 instead of 6/7.
What is in the open slice, and what is not
The line is drawn so that the open tables are an index into the dataset rather than a copy of part of it. Both carry the conditions and the scores; neither carries a number we had to form a judgment to produce.
- Tripwire ids, conditions as written, status, and the note on each
- Which scenario each condition belongs to, and whether that scenario is a root trajectory or a branch
- Drill run ids, dates, which drill, how it was played, and the reached-of-total score
- Probability ranges, their full revision history and the written reason behind every move
- The daily signal ledger and the source annotations
- Scenario narratives, preconditions, leading indicators and the Prevent · Detect · Respond · Recover playbooks
- Drill transcripts, decision points, the checks themselves and their evidence
- The Standard, the trajectory forces and the readiness instrument
See Data licensing.
How to credit it
CC BY asks for attribution and a link. Either of these is fine:
Tripwire data from AGI Watch, CC BY 4.0.
Bhutani, V. & AGI Watch (2026). AGI Watch tripwire board [Data set]. https://agiwatch.ai/open-data. CC BY 4.0.
Two things CC BY does not let you do, and both matter here: do not imply AGI Watch endorses your work or conclusions, and do not present a stale copy as current. The tables are regenerated on every deploy and each carries a generated date in the JSON — quote it, or link the live file.
Why only two tables
Because a citable table is only worth publishing if it is the thing nobody else has. Both of these are, as far as we can establish: no other public site maintains a dated board of falsifiable scenario conditions with a tripped-or-clear status — the idea exists in the literature as one-off papers with no status attached — and none publishes scores from AI-scenario tabletop exercises. The organisations that run such exercises publish findings, not results.
Widening the slice is easy and narrowing it after the fact is not, so it starts here. If you need something adjacent — the histories behind a move, the signals, a feed — say what for: [email protected].
Put the board on your own site
The tripwire board is embeddable. It renders the conditions that have crossed, grouped by branch, in dark or light to match the page around it, and it carries the attribution CC BY asks for so dropping it in is already compliant. One iframe, no script, nothing tracked:
<iframe src="https://agiwatch.ai/embed/tripwires"
title="AGI Watch — tripwires crossed"
width="100%" height="520" loading="lazy"
style="border:1px solid #1d2230;border-radius:10px"></iframe> It rebuilds whenever the board changes, so an embed never goes stale on its own. Set the height to suit your layout — it scrolls rather than clipping. If you would rather build your own chart, take the CSV.
Robots
Everything under /open-data is explicitly allowed to every crawler, including the model-training crawlers that are refused on the rest of the site. That is the point of releasing it under CC BY 4.0: an open table nobody may fetch is not open. The reasoning behind the split for the rest of the site is in the methodology and the terms.
Attribution string: AGI Watch (agiwatch.ai), method by Vibhu Bhutani