AI lab researchers and engineers on X — evidence-linked accounts

For the 19 resolver-backed labs, no: the candidate roster comes from OpenAlex, arXiv, Wikidata, and public GitHub membership. For the 13 newer groups, bounded X or stored-dataset queries can surface candidate leads, alongside official team pages, model/report bylines, and other primary professional sources. A search hit or bio keyword is never enough to publish: exact account identity is checked independently, and current lab affiliation is checked against the current bio plus primary professional context. The separately labeled maintainer overrides have their own public-record path. Targeted groups were not run through the structured resolver, so their source pools are shown as "Not sampled."

101 listings on this page (out of 236 people) use bio_self_declared: the person's own current X bio explicitly states the assigned affiliation. This is not a blind bio-keyword sweep — every person was identified independently first. Some belong to the structured labs' academic rosters; 58 were identified through primary professional evidence, including official team pages and model or report bylines, for the 13 targeted groups. If a later bio drops the lab wording, silence alone is not a contradiction; material ambiguity is investigated with current primary sources, and a listing is removed when evidence contradicts the assignment. Bio text remains lower confidence than a structured identity-registry field, so the label stays visible in the cards, table, and CSV.

3 listings — Sam Altman, Andrej Karpathy, and Timothée Lacroix — are retained through current public records rather than a structured or current-bio affiliation path. Altman's current bio quotes OpenAI's mission and carries X's OpenAI affiliate badge, while OpenAI's own governance page names him CEO; Karpathy's bio is empty, but his first-person announcement and current reporting place him on Anthropic's pre-training team; Lacroix's bio still reads "PhD student at Facebook AI Research Paris," while Mistral's current company page names him co-founder and CTO. These three were added at the site maintainer's explicit direction. This is the least rigorous tier on the page by design, and it is labeled "Public record (override)" everywhere.

No, and it undercounts substantially on purpose — this is a floor, not a census. Four OpenAlex institution rosters are capped at a 120-person sample rather than a full population. Some older published OpenAlex and GitHub provenance does not belong to the current sampled pool, so an overall conversion percentage would mix cohorts; the source sizes and signal-listing counts are shown separately. The comparable Wikidata-employer cohort resolved 24% to a self-asserted X handle. The 13 newer groups have no structured denominator; their official pages and technical bylines are bounded discovery slices. Most working researchers never link an X account machine-readably, and official named rosters cover only a slice of each organization.

The exact account is linked to an independently identified person through an ORCID researcher-url, Wikidata P2002 claim, GitHub profile field, self-maintained professional page, or clearly labeled public-record override. Those handle links establish identity, not employment. Current lab affiliation is reviewed separately against the public X bio and current primary professional context; a current self-authored professional profile can support a bio-silent case. The 58 listings in newer groups use the lower-confidence bio tier after independent identity checks. Established professional aliases stay as published identities without an invented legal name. A name match, follower graph, or parent-company-only claim is insufficient; material ambiguity stays under review.

The last global live-endpoint and bio-contradiction sweep was October 20, 2026. It reviewed all 227 existing lab listings (226 distinct handles): 222 listings returned live profiles and five unavailable handles returned 404, so their prior entries remain visible without new enrichment. Newly identified people were also checked against current public X profiles and independent identity sources. Current-tense affiliation supports an assignment; explicit past-tense wording or a different current employer triggers removal or further investigation. Silence or unresolved subteam scope alone is not treated as a departure. This remains a point-in-time snapshot, not a live feed.

Qwen, Tencent Hunyuan, ByteDance Seed, and Xiaomi MiMo are model teams inside much larger companies rather than their own OpenAlex institutions, and using the parent company as the roster source would reproduce the exact false-positive problem this methodology exists to avoid. MiniMax does have its own Wikidata organization and a real technical report, so its roster is pinned to that exact OpenAlex work instead of a broad affiliation-text search; the one resolved handle belonged to someone whose current bio names different employers with zero MiniMax mention, so it was excluded. The current configured source pools for these five teams produced no ORCID, Wikidata, or GitHub handle link that survived the employment check. Tencent Hunyuan, ByteDance Seed, and Xiaomi MiMo show resolved people anyway because a follow-up pass individually identified specific researchers, several already inside the sampled academic rosters, whose own X bio stated the team affiliation — see the "Bio (lower confidence)" FAQ. Qwen and MiniMax haven't had that same follow-up pass complete yet, so zero is a coverage gap rather than a confirmed true negative.

Yes — the full roster (name, lab, X handle, source signal, followers, verified, role, previous affiliations) is published as a CSV under CC BY 4.0 at https://crawlora.net/ai-researchers-index/download. No email addresses or other non-public contact info are collected or published.

Same source as everything else self-declared on this page: the person's own X bio. Where a bio states a current title ("Research Engineer", "CEO & Co-Founder") we surface it near-verbatim in "Role"; where it uses explicit past-tense language ("previously/prev/formerly/ex-") naming a prior employer, we surface that in "Previously at". Unlike the lab-affiliation signal (ORCID, Wikidata, GitHub — all structured records), these two fields are extracted from free-text bio content, so they're held to a stricter bar: ambiguous, terse, or unclear bio text is left blank rather than guessed. A blank cell means the bio didn't clearly state one, not that we don't know.

That earlier study measured how often AI-lab names appear as text inside bios across the X Users dataset — an aggregate keyword count with no named individuals and no employment verification. This page starts from named people in academic, organizational, or official product/research rosters (OpenAlex, arXiv, Wikidata, GitHub, official team pages, and model/report bylines), then pairs each with a labeled identity and affiliation path.