This ODRL Profile defines AI-specific actions which enables policy authors to clearly define acceptable and prohibited uses of assets in AI workflows. It supports governance, risk management, and compliance efforts, helping ensure that AI systems operate within agreed legal, ethical, and contractual boundaries.

Introduction

This ODRL Profile defines a controlled vocabulary of actions governing the use of assets by Artificial Intelligence (AI) systems. As AI technologies increasingly interact with digital and physical assets—through training, inference, transformation, and decision-making—there is a need for precise, machine-readable policies that express permissions, prohibitions, and obligations specific to AI-driven activities.

This ODRL Profile addresses this gap by defining a set of actions tailored to AI systems. These include permissions and prohibitions related to training models, generating outputs, modifying assets, and redistributing AI-derived content. By enabling explicit and machine-readable policy statements, this profile supports responsible AI practices and facilitates enforcement of organisational, legal, and societal expectations.

Identifiers

The Profile URI for this vocabulary is defined as:

Within this document, the following namespace prefix bindings are used:

This Profile uses wording according to [[!RFC2119]].

AI "opt-out" Vocabularies

Numerous standards groups and other projects have developed vocabularies for "opt-out" of AI processes over user content. An analysis has found them to be divergent, fragmented, and imbalanced and also provides a detailed summary of these vocabulary options.

Below is a summary of the key proposals.

IETF - AI Preferences [[?draft-ietf-aipref-vocab]]

• Foundation Model Production (aitrain) - The act of using an asset to train or fine-tune a foundation model
• Search (search) - using one or more assets in a search application that directs users to the location from which the assets were retrieved

Creator Assertions Working Group (CAWG) - Training and Data Mining Assertion [[?cwag-train]]

• cawg.data_mining - Can any text or data content be extracted from the asset for purposes of determining “patterns, trends, and correlations.”
• cawg.ai_inference - Can the asset be used as input to a trained AI/ML model for the purposes of inferring a result
• cawg.ai_generative_training - Can the asset be used as training data to an AI/ML model that could generate assets.
• cawg.ai_training - Can the asset be used as data to train non-generative AI/ML models, such as those used for classification, object detection, etc.

PLUS License Data Format - Data Mining [[?plus-data-mining]]

• DMI-ALLOWED - Data Mining allowed
• DMI-PROHIBITED - Data Mining prohibited
• DMI-PROHIBITED-AIMLTRAINING - Prohibited for AI/ML training
• DMI-PROHIBITED-GENAIMLTRAINING - Prohibited for Generative AI/ML training
• DMI-PROHIBITED-EXCEPTSEARCHENGINEINDEXING - Prohibited except for search engine indexing

Really Simple Licensing (RSL) - AI Usage [[?rsl-ai]]

• all - Any automated processing, including AI training and search
• ai-all - Any use by AI systems, including training, inference, indexing, retrieval, generation, grounding, or model evaluation. This term explicitly includes the ai-train, ai-input, and ai-index usage types defined below, as well as other AI-related uses not yet enumerated
• ai-train - Training or fine-tuning AI models
• ai-input - Input into AI models, including retrieval-augmented generation, grounding, or other use of the content to produce generative AI responses or search summaries
• ai-index - Inclusion in an AI system’s internal index or retrieval database
• search - Building a search index and providing search results (e.g., returning hyperlinks and short excerpts from your website's contents). Search does not include providing AI-generated search summaries

As can be clearly seen by these existing approaches, they are all inconsistent in semantics, and hence lack data interoperability for AI governance.

ODRL AI Vocabulary Profile

Using the current set of AI vocabularies described in Section 3 as a snapshot of the AI community requirements, the ODRL AI Vocabulary Profile includes the following semantic terms:

• Data Mining - any form of algorithmic analysis of an asset
• AI Training - the asset is used to train an AI model
• Generative AI Training - the asset is used to train a Generative AI model
• AI Augmentation - the asset is used as input to a trained AI model
• Index - the asset is registered in a data store
• Search - the asset is discoverable in a data store

These vocabulary terms are shown below, including their hierarchy.

The next sections formally defines these vocabulary terms.

Examples

The asset (dataset1) maybe used for AI Training only for the explicit purpose of "Research and Development"

{
  "@context": "http://www.w3.org/ns/odrl.jsonld",
  "@type": "odrl:Set",
  "profile": "https://w3c.github.io/odrl/ai-vocab/",
  "uid": "http://example.org/policy/001",
  "target": "http://example.org/asset/dataset1",
  "permission": [{
      "action": "odrl:ai-training",
      "constraint": [{
          "leftOperand": "odrl:purpose",
          "operator": "odrl:eq",
          "rightOperand": "https://w3id.org/dpv#ResearchAndDevelopment"
       }]
    }]
}

The asset (dataset2) maybe used for Data Mining but must not be used for AI Augmentation. Note that the conflict property is set to prohibit to ensure that is overridden in the Data Mining action hierarchy.

{
  "@context": "http://www.w3.org/ns/odrl.jsonld",
  "@type": "odrl:Set",
  "profile": "https://w3c.github.io/odrl/ai-vocab/",
  "uid": "http://example.org/policy/002",
  "conflict": "prohibit",
  "target": "http://example.org/asset/dataset2",
  "permission": [{
        "action": "odrl:data-mining"
    }],
  "prohibition": [{
      "action": "odrl:ai-augment"
    }]
}