Who actually shapes AI policy in the U.S.? We mapped 1,812 entities: 745 people, 918 organizations, 2,925 relationships. Frontier Labs, AI Safety orgs, Think Tanks, Government, VCs, and more. mapping-ai.org

Thanks so much @AndyMasley (and others @ohlennart @PeterSlattery1 @austinc3301 @tobias_haeberli @DrTechlash etc) for your careful and critical engagement, we’re genuinely grateful. On the errors you flagged (understood these are a limited sample), we've corrected all of them. We’ve also made bidirectional edge relationships more legible. Noting here that edges have start/end dates in our database, but we haven't shipped the "current only" filter. That's coming soon and will fix the John Schulman-type issues structurally. We’ve been manually verifying samples, running verification scripts 24/7, and have worked with a dedicated data validation contractor in the past. Though at this scale, errors are inevitable, and it’s thanks to exploration like yours that we’re able to strengthen our verification methodology. This is also why we’ve made our full codebase open-source on GitHub (github.com/MappingAI/mapping…) and host a forum on our Discord (discord.gg/HtqceQRV3f) for deeper and collaborative debugging. We welcome you and any interested community members to join one or both. You’re right that we should do a better job at publishing a formal methodology document, that is next on our task list. The GitHub repo itself and our mapping party guide (mapping-ai.org/workshop) give pointers to the architecture and relevant files, and on the site we’ve included a note that data comes from a mix of public records, user submissions, and LLM-assisted research. We’ve also stated that inferred stances don’t claim to represent official positions. There's also more on our process on "About" and added as tooltips across the site. On taking the site down, we take this feedback seriously, and we’ve thought hard about it. Keeping this tool online, with the right precautions, offers us the public exposure to surface issues and seed data at a scale far larger than our team’s capacity (mostly just two of us with full-time day jobs). See Cunningham’s Law (meta.wikimedia.org/wiki/Cunn…). Which is also why we’ve done three things immediately (see screenshots): 1) There is a persistent pop-up that must be dismissed by the user on every page load of the site, regardless of local session cache, that makes it clear to both insiders and outsiders coming to this tool that errors exist and are being actively reviewed and addressed, as well as additional notes in our footer and “How to Use” pop-ups. 2) Org stances are temporarily removed (thank you @ohlennart for this suggestion as well). Until we’ve internally verified every source and edge type, we’re pulling stance attributions. Updated tooltips on the legend briefly explain our methods and this temporary removal for orgs. 3) Every field on every entity now has verification flags (green/yellow/red), upvote/downvote buttons to confirm or flag data, and a rich-text correction notes feature with @mentions. These update live, and automated verification scripts cross-check fields against external sources. A "Data Quality" legend on the map lets users filter by verification status. Together, these tools let the community and our team collectively triage and fix data quality issues in real time, without misrepresenting accuracy. More broadly on org stances: although many orgs don’t have official positions (whether by law or org policy), our goal is to capture what’s being discussed publicly about organizations in the AI policy space, including external and internal perceptions alongside officially stated positions. We see the gap between internal and external perceptions as informative rather than something to flatten. And we include “inferred” vs “explicitly stated” tags to help draw that distinction for users. That said, we respect organizational corrections and will update promptly when groups reach out, as we’ve done here and with several other organizations since launch. And thanks to those who have reached out with corrections and feedback via email, Discord, and the contribute form! We’re actively building the infrastructure to make this kind of collective intelligence possible. Thank you for making this tool better and for your patience along the way :) (and if you made it this far, thanks for reading!) x.com/mapping_ai/status/2051…

shout out to @JustenMichel for his AI tribes post that we referenced in first trying to make sense of this space! substack.com/home/post/p-185… would love any feedback you have :) or a collab? 👀