Open model
A model with openly available code or weights.
How the laws define this
Both instruments agree that open models are distinct from proprietary approaches because they facilitate access to essential resources for development, training, or evaluation rather than remaining entirely closed. They diverge, however, on whether openness applies to the output artifacts of the model or to the input data used to build it, as well as the degree of restriction imposed on that access. The Responsible AI Guidance for the Public Service: GenAI focuses on the openness of the model itself, defining open models by the public availability of internal artifacts—such as code, weights, and documentation on training data and provenance—to enable inspection, reuse, and external validation. In contrast, the Draft AI Act presented by political party 'Da, Bulgaria' (project bill for an AI law) defines openness through controlled access to public-sector archives and datasets. Furthermore, the draft bill from 'Da, Bulgaria' restricts training and use specifically to non-commercial and research contexts, enforcing explicit safeguards like privacy rules, licensing, and access controls to support local development while preventing misuse.
Synthesised from the 2 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| New Zealand | Responsible AI Guidance for the Public Service: GenAI | In Force | This definition expands openness beyond code or weights to include documentation about training data and provenance, emphasizing facilitation of inspection, reuse, and validation. |
| Bulgaria | Draft AI Act presented by political party 'Da, Bulgaria' | Draft | This definition redefines openness as access to public archives and datasets for non-commercial and research use, with specified safeguards like privacy and licensing. |
What it triggers
Definitions (2)
An 'open' model is distinguished from proprietary models by greater transparency of model artifacts—such as publicly available code, weights, or documentation about training data and provenance—facilitating inspection, reuse and external validation as outlined in the guidance.
A model characterized by provisions for controlled access to public archives and datasets for training and use in non-commercial and research contexts; the draft defines conditions and safeguards (privacy, licensing, access controls) for using public-sector data to support local open-model development while restricting misuse.
Not to be confused with
Open model offers greater transparency and access compared to proprietary model, which is controlled under restricted terms by a vendor.
Open model encompasses broader transparency aspects beyond just publicly available weights.
Open model specifies access and transparency features, whereas generative AI (GenAI) refers to content-generating capabilities.
Questions and answers
What makes a model 'open' under New Zealand guidance?
It features public availability of code, weights, or documentation on training data and provenance for inspection and reuse.
How does Bulgaria's draft define open models?
It ties openness to controlled access to public data for non-commercial research with privacy and licensing safeguards.
How does open model differ from proprietary model?
Open models provide transparency and access, while proprietary models restrict access under vendor control.
Related Terms
Proprietary model
A model owned and controlled by a vendor under restricted terms....
Open-weight model
A model whose weights are publicly available....
Open models
Models using controlled access to public datasets for reuse....
Model
A trained algorithmic artifact that performs tasks using learned parameters....
Generative AI (GenAI)
AI that produces new content like text, images or code....