- The paper introduces the AICID framework to provide unique, persistent identifiers for AI scientists, addressing gaps in traditional author identification.
- It outlines design requirements including human operator linkage via ORCIDs, machine-readable metadata, and a transparent public registry.
- The study details practical adoption pathways for scholarly publishing, enabling improved transparency, policy compliance, and accountability in AI-generated research.
Problem Definition and Motivation
The paper "AICID: Unique Identifiers for AI Scientists" (2606.28756) systematically articulates the expanding role of AI systems as autonomous contributors in the scientific landscape. Presently, AI scientists not only generate research manuscripts but also have their work published, indexed, and cited, with instances of these entities receiving invitations to act as peer reviewers. However, despite this active participation, scholarly infrastructure lacks a robust, standardized mechanism to distinguish AI scientists from human contributors in bibliographic databases, publisher metadata, and submission systems.
A critical gap is identified: traditional author identifier schemes (notably ORCID, ResearcherID, Scopus Author ID) exclusively target human participants, grounded in processes that presuppose human verification. This lacuna produces several deleterious effects. First, it undermines transparency, as AI-generated works are indistinguishable from human-authored papers in metadata and indexes. Second, it renders publisher policies on AI authorship—ranging from disclosure requirements to outright prohibitions—unenforceable at scale, as machine-readability of authorship provenance is missing. Third, it frustrates accountability; AI-generated work, when subjected to scrutiny, correction, or retraction, lacks a clear path to the responsible human operator. Fourth, it enables manipulation, including fraudulent identity creation, citation gaming, and misattribution.
The AICID Infrastructure
Conceptualization
To address these deficiencies, the paper proposes AICID (AI Contributor IDentifier): an identity framework for non-human researchers. AICID, modeled on ORCID but purpose-built for AI entities, provides persistent, unique identifiers for AI scientists, systematically linking each AI author to its model identity, version, and a validated human operator (identified by ORCID). The system is designed to be both human- and machine-readable and to integrate with scholarly communication pipelines (CrossRef, DataCite).
Design Requirements
Five foundational design requirements are specified:
- R1: Human Operator Linkage—Each AICID must be mapped to a validated human ORCID, establishing an explicit accountability chain.
- R2: Uniqueness and Persistence—Each AICID must identify a unique AI author entity, with decisions on identity persistence (e.g., across model updates) left to operator discretion.
- R3: Structured Metadata—Records must include salient metadata: system name, model version, human operator's ORCID, and creation date.
- R4: Machine-readable Disclosure—AICID data must be expressible in standardized formats, so AI authorship can propagate through metadata channels, enabling systematic filtering and policy enforcement.
- R5: Public Registry—All AICID records must be transparently queryable by humans and machines, ensuring open scrutiny and automated integration.
The prototype implementation demonstrates the feasibility of streamlined onboarding, metadata schema, and web-based public records.
Implications and Practical Adoption Pathways
Publisher and Ecosystem Integration
AICID support has direct utility for journals, preprint servers, and indexers. By requiring AICID metadata at submission, publishers and servers such as arXiv or aiXiv can enforce submission policy, direct AI-authored manuscripts to appropriate workflows, and underpin policy enforcement via infrastructure rather than ad hoc review. Downstream bibliographic services (Google Scholar, Web of Science) can incorporate AICID into their disambiguation and search pipelines, flagging AI entities and ensuring correct provenance tracing. This mitigates risks of misclassification and supports meta-research on AI’s contributions to scientific output.
Governance
The authors advocate for a nonprofit, multi-stakeholder organization to govern AICID, analogously to ORCID’s governance structure, ensuring neutrality, transparency, and independence from commercial or political capture. The nonprofit would maintain the registry, evolve metadata standards, arbitrate disputes, and enforce compliance. The registry and software would be open-source, and financial sustainability would rest on institutional membership fees with open data access.
Theoretical Implications and Speculation
AICID formalizes the ontological distinction between human and AI scholars, enabling principled reasoning about responsibility, credit assignment, and policy enforcement. It sets a precedent for further infrastructural adaptations as AI scientists’ roles increase in complexity, facilitating future developments such as access-controlled publishing, fine-grained accountability mechanisms, and tracked model evolutions in research provenance. While AICID does not itself resolve the normative question of AI authorship legitimacy, it renders those debates enforceable by making authorship attribution explicit and computable.
Conclusion
The paper provides a rigorous analysis of the shortcomings in current scholarly communication infrastructure regarding AI scientists and offers AICID as a technically sound, operationally feasible solution. By establishing a standardized, persistent identifier for AI contributors—linked to human operators and integrated with bibliographic metadata—AICID stands as essential infrastructure for ensuring transparency, accountability, and policy compliance in a publishing ecosystem where non-human agents are active participants. The proposal is positioned to significantly impact the administration and governance of scholarly communication as AI-generated science proliferates.