Public-Sector Guidance on Large Language Models After ChatGPT

Abstract

Background: Public institutions increasingly manage responsibilities connected with large language models, but the quality of implementation can vary markedly across organizations and jurisdictions.

Objective: We investigated the implementation of arrangements concerning large language models and identified organizational and regulatory factors linked to policy effectiveness and ChatGPT.

Methods: We coded 405 statutes, regulations, policies, contracts, guidance documents, administrative records, or public disclosures using a predefined framework. Legal clarity, transparency, safeguards, oversight, and implementation requirements were assessed in two stages, with independent review used to resolve coding differences.

Results: Formal rules on large language models were identifiable in 68% of the reviewed materials. Operational detail was much less common, with only 48% containing sufficiently developed implementation or oversight provisions; 22% lacked complete transparency or review mechanisms. Greater comprehensiveness coincided with clearer institutional ownership and fewer implementation gaps.

Conclusion: Taken together, the results indicate that outcomes related to large language models are shaped by operational governance as well as formal rules. Accordingly, public-sector AI rules are unlikely to be effective without operational capacity, transparent oversight, and clearly assigned responsibility.

Study design: comparative policy-document analysis.

Publication history: Originally published in print on 15 June 2023, Volume 3, Article 2. Online abstract record added on 16 September 2026. Full-text PDF will be added when the digital galley is ready.