Instrument, seed build · seed-2026-09-12
AI Risk Coverage Monitor
A live index of AI risk coverage in every language, measured against its primary sources. Measurements, records, and a workbench where you combine them. The site states what documents say and when. It never states why.
191
records in the seed corpus
12
days, Sep 1 to 12, 2026
1 of 65
languages tracked (seed is English; live is global)
11 of 12
largest headline word gap in the seed
News items per day, 71 dated of 83Podcast episodes and broadcast segments per day, 37Peak: Sep 10, 2026, 16 items
Live wire
09-11ForecastStated probability: 30 to 60 percent odds of a fatal AI disaster09-11Personnel changeNBC reports two further researcher departures citing safety concerns09-11Research publicationAI legislative update, September 11, 202609-10Safety disclosureAnthropic publishes Detecting and countering misuse of AI: September 202609-10Government actionSenator Hawley opens a subcommittee investigation into OpenAI, documents due October 109-10Open letterOpen letter about the Mathathon, 771 signatories09-10ForecastScienceSoft forecasts AI risk in 60 to 80 percent of new liability and cyber underwriting by 202809-10Advocacy actionARI launches a 50 state advocacy initiative09-09Government actionGovernor signs AB 1405 and SB 81309-09Safety disclosureAnthropic publishes an alignment assessment of recent cybersecurity incidents09-09Personnel changeResearcher Jacob Coxon resigns and publishes a statement09-09Personnel changePaul Christiano joins the OpenAI Foundation board as a non voting observer on the Safety Committee09-09ForecastStated probability: above 10 percent within the decade09-09ForecastStated probability: about 4 percent within one year and 15 percent within three years09-11ForecastStated probability: 30 to 60 percent odds of a fatal AI disaster09-11Personnel changeNBC reports two further researcher departures citing safety concerns09-11Research publicationAI legislative update, September 11, 202609-10Safety disclosureAnthropic publishes Detecting and countering misuse of AI: September 202609-10Government actionSenator Hawley opens a subcommittee investigation into OpenAI, documents due October 109-10Open letterOpen letter about the Mathathon, 771 signatories09-10ForecastScienceSoft forecasts AI risk in 60 to 80 percent of new liability and cyber underwriting by 202809-10Advocacy actionARI launches a 50 state advocacy initiative09-09Government actionGovernor signs AB 1405 and SB 81309-09Safety disclosureAnthropic publishes an alignment assessment of recent cybersecurity incidents09-09Personnel changeResearcher Jacob Coxon resigns and publishes a statement09-09Personnel changePaul Christiano joins the OpenAI Foundation board as a non voting observer on the Safety Committee09-09ForecastStated probability: above 10 percent within the decade09-09ForecastStated probability: about 4 percent within one year and 15 percent within three years
01 · Headline versus source
The cited passage beside every headline that cites it
One primary document. Its verbatim passage. Every headline recorded as citing it, word by word. The build counts the words absent from the passage; the reader reads the rest. 3 pairs, 6 queued.
Primary source, the cited passage
could support weapons development
AnthropicThreatReport:AIModelsNearBioweaponsThresholdasDroneKillSoftwareEmerges
11 of 12No qualifying verb found
AnthropicsaysscientistsexploitedClaudeforbioweaponsresearch
5 of 6No qualifying verb found
justthenews.com phrase as recorded
AnthropicrevealsAIusedtotestmissiles
6 of 6No qualifying verb found
filled headline word absent from the passage · outlined word present in the passage · plain stop word, not counted. The meter is absent words over counted words. Computed by the published rules; nothing else is attached.
02 · The record
Every dated document, January to September 12, 2026
49 primary source records and 21 public filings dated 2026, one dot each, in four lanes. Hover any dot. No two dots are connected by the site; the Workbench lets you connect what you choose.
Labs: releases and disclosuresGovernment: actions and billsMoney: rounds, filings, contributionsPeople and papers: departures, letters, research, forecasts
03 · The economics of AI risk
7 forecasts of the same market, 183 times apart
Analyst forecasts for the AI governance, assurance, and red teaming markets, each drawn from its base year to its target year on a log scale. The disagreement is the finding. Philanthropy, institute budgets, and insurance figures sit in the same dataset.
Forecast, base to target yearHighest forecast in the setLog scale. Largest target over smallest: 183x
04 · Public records
Contributions and lobbying, as filed
FEC and Senate LDA records, dated, each bar a link to the filing or committee page it is derived from. A layer you can place on the timeline yourself. The site places it beside nothing.
Contributions, FEC derived, through June 30, 2026 (5 rows)
Federal lobbying, Senate LDA, Q1 and Q2 2026 (11 rows)
Q1 2026Q2 2026
06 · Discourse
37 episodes and segments in twelve days
Podcast episodes and broadcast segments on AI risk, by day and by show. Show, title, host, guest, date, link. No column describes anyone's position. All in the Workbench.
014
022
033
045
051
061
070
081
094
104
117
125
05 · What the words say
Verb class and source type across the seed
The keyword rule applied to 83 headlines, and the public registry applied to their domains. Both rules are on the Methods page and any reader can re-run them.
No qualifying verb found 79
Conditional language 4
Source type by domain registry, 83 items
09 · Forecasts, not facts
Stated probabilities and market prices
Individuals' stated probabilities as published, and prediction market prices when live: Metaculus (reputation points), Polymarket (crypto settled), Kalshi (CFTC regulated). The settlement type changes what a price means, so it is disclosed.
Sep 10, 2026
Sep 9, 2026
Stated probability: above 10 percent within the decadeIndividual, Evan Hubinger
Sep 9, 2026
Stated probability: about 4 percent within one year and 15 percent within three yearsIndividual, Paul Christiano
Sep 11, 2026
Stated probability: 30 to 60 percent odds of a fatal AI disasterIndividual, Greg Jensen
Sep 9, 2026
Prediction markets
awaiting first fetch
07 · Coverage Intensity Index
Four component series, one reading, published weights
Each series is normalized to its own 90 day median. The reading is undefined until 30 days of baseline exist, and the ring says how far along that is rather than painting a number.
Baseline collecting. The reading appears at day 30, not before.
Weights published. Set your own in the Workbench.
Coverage volume w 0.4
no data yetGDELT DOC, all languages
Broadcast attention w 0.2
no data yetGDELT TV Explorer
Discourse volume w 0.2
no data yetHacker News, Wikipedia, podcasts
Front page prominence w 0.2
no data yetGDELT Frontpage Graph
08 · Revisions
What changed in an article after it went up
From GDELT's Global Difference Graph, a before and after of the title at 24 hours and one week. Never a verdict on why.
Status
no revisions observed yet
The Difference Graph feed starts with live ingestion.
10 · Notes
Reader notes, ranked by people who usually disagree
Any exhibit, record, or pair takes a note. A note needs a source URL and 280 characters. It shows immediately under Needs more ratings and is promoted only when raters from different rating histories both find it helpful. Pseudonyms only; no email, no name, no IP. How notes work.
Identity
random pseudonym
Issued in a cookie. Nothing else is stored.
Rank rule
notes rank v1
5 ratings, 70 percent helpful, 2 rater clusters.