Public actor registry

Who shapes Germany’s public AI ecosystem?

This registry connects organizations, roles and evidence across funding, research, procurement, influence and infrastructure. A connection counts only with an official identifier or an explicit identifier crosswalk—matching names alone never suffice.

Every claim has provenanceNo name-based mergesNo generative AI model
Latest source snapshot 17 Jul 2026, 05:206 public projectionsSchema ai-actor-registry/v1
separate actor records1,347bounded by official or source-scoped identifiers
evidence-backed relations2,615each tied to a dataset and source record
confirmed cross-lane merges0only an explicit identifier crosswalk may increase this value
name pairs for review11not merged and not presented as confirmed identity
Critical boundary

The same name does not mean the same organization

An exact normalized full name creates a review candidate only. Both records, identifiers and roles stay separate until a reliable identifier crosswalk exists.

mergeApplied = false
1,347 results
Actor records

Choose an organization

40 of 1,347 visible
Selected actor

TÜV-Verband e.V.

actor:bundestag-lobbyregister:R000008
officially identified
Label state
official labels are consistent
Data lanes
Influence & lobbying register
observed roles
registered lobby organization
Evidence claims
1

Identifiers

  • bundestag-lobbyregisterR000008

Source evidence

AI influence & public money1 source records · org:lobby:r000008

Evidence-backed roles and relations

  • Influence & lobbying registerregistered as an AI-relevant lobby actorAI influence & public money
Coverage & gaps

What each data lane can contribute to identity

The lanes have different identifier quality. Source-scoped roles are never connected automatically to officially identified organizations; the patent lane stays blocked while its public projection lacks an applicant identifier.

Identity coverage of the public AI actor registry by data lane
Data laneIdentity statusInput recordsActorsClaimsUnresolved
Fundingofficial identifiers only2,1278812,1270
Research & knowledge flowofficial identifiers only3003003000
Public procurementofficial identifiers only1110110
Influence & lobbying registerbounded official-identifier view1531531740
Infrastructuresource-scoped identities only3330
Patentsno actor identifiers availableThe public patent projection contains family and geography signals but no applicant identifier.8,241000
Method & guardrail

Identity without invented bridges

The registry is a deterministic projection of six public datasets. It groups roles around stable identifiers and shows uncertain name pairs separately.

  1. 1
    Take official identifiers

    PIC, ROR, OpenAlex, procurement and lobby-register identifiers define separate actor records.

  2. 2
    Merge only explicit identifier crosswalks

    Several identifiers enter one actor only when a single source record explicitly provides that crosswalk.

  3. 3
    Use names as a review signal only

    An exactly normalized full name creates a candidate with mergeApplied=false, never an automatic identity.

  4. 4
    Keep gaps visible

    When an organization identifier is missing, a role stays source-scoped or the whole lane remains visibly blocked.

Collection, identity logic, linking and presentation use no generative AI model.

Source snapshots

Six projections, separate timestamps

The latest source snapshot shown above does not replace individual dates. Here you can see when each published projection was generated.

  • AI patent families with German association signals16 Jul 2026, 15:58
  • AI influence & public money17 Jul 2026, 05:20
  • CORDIS funding & innovation path17 Jul 2026, 05:20
  • OpenAIRE/OpenAlex knowledge flow16 Jul 2026, 22:16
  • public AI infrastructure13 Jul 2026, 20:47
  • AI procurement supplier index15 Jul 2026, 16:18

Frequently asked questions

Is every entry a uniquely confirmed legal entity?
Officially identified entries are based on a published identifier. Source-scoped infrastructure roles, by contrast, are bounded only within their source dataset and are not merged across lanes.
Why do familiar organizations appear more than once?
Different registers use different identifiers. Without an explicit identifier crosswalk, combining them would be a guess, so the records stay separate.
Are the name pairs confirmed matches?
No. They are a review queue only. Every candidate has state review_required and mergeApplied=false.
Why are patent applicants missing?
The current public patent projection contains family, time and geography signals but no reliable public applicant identifier. That lane therefore remains visibly blocked.

Want to understand public AI actors across data lanes?

We build traceable registries, collectors and evidence graphs with conservative identity rules and verifiable provenance.

Discuss a data projectView all tools