A Vocabulary For Expressing AI Usage Preferences
draft-ietf-aipref-vocab-02
| Document | Type | Active Internet-Draft (aipref WG) | |
|---|---|---|---|
| Authors | Paul Keller , Martin Thomson | ||
| Last updated | 2025-07-21 | ||
| Replaces | draft-keller-aipref-vocab | ||
| RFC stream | Internet Engineering Task Force (IETF) | ||
| Intended RFC status | Proposed Standard | ||
| Formats | |||
| Additional resources | Mailing list discussion | ||
| Stream | WG state | WG Document | |
| Document shepherd | (None) | ||
| IESG | IESG state | I-D Exists | |
| Consensus boilerplate | Yes | ||
| Telechat date | (None) | ||
| Responsible AD | (None) | ||
| Send notices to | (None) |
draft-ietf-aipref-vocab-02
AI Preferences P. Keller
Internet-Draft Open Future
Intended status: Standards Track M. Thomson, Ed.
Expires: 22 January 2026 Mozilla
21 July 2025
A Vocabulary For Expressing AI Usage Preferences
draft-ietf-aipref-vocab-02
Abstract
This document proposes a standardized vocabulary for expressing
preferences related to how digital assets are used by automated
processing systems. This vocabulary allows for the creation of
structured declarations about restrictions or permissions for use of
digital assets by such systems.
About This Document
This note is to be removed before publishing as an RFC.
The latest revision of this draft can be found at https://ietf-wg-
aipref.github.io/drafts/draft-ietf-aipref-vocab.html. Status
information for this document may be found at
https://datatracker.ietf.org/doc/draft-ietf-aipref-vocab/.
Discussion of this document takes place on the AI Preferences Working
Group mailing list (mailto:ai-control@ietf.org), which is archived at
https://mailarchive.ietf.org/arch/browse/ai-control/. Subscribe at
https://www.ietf.org/mailman/listinfo/ai-control/.
Source for this draft and an issue tracker can be found at
https://github.com/ietf-wg-aipref/drafts.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Keller & Thomson Expires 22 January 2026 [Page 1]
Internet-Draft AI Preference Vocabulary July 2025
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 22 January 2026.
Copyright Notice
Copyright (c) 2025 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Conventions and Definitions . . . . . . . . . . . . . . . . . 3
3. Statements of Preference . . . . . . . . . . . . . . . . . . 4
3.1. Conformance . . . . . . . . . . . . . . . . . . . . . . . 5
4. Vocabulary Definition . . . . . . . . . . . . . . . . . . . . 5
4.1. Automated Processing Category . . . . . . . . . . . . . . 6
4.2. AI Training Category . . . . . . . . . . . . . . . . . . 6
4.3. Generative AI Training Category . . . . . . . . . . . . . 7
4.4. AI Use Category . . . . . . . . . . . . . . . . . . . . . 7
4.5. Search Category . . . . . . . . . . . . . . . . . . . . . 7
5. Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.1. More Specific Instructions . . . . . . . . . . . . . . . 7
5.2. Vocabulary Extensions . . . . . . . . . . . . . . . . . . 8
6. Exemplary Serialization Format . . . . . . . . . . . . . . . 8
6.1. Usage Category Labels . . . . . . . . . . . . . . . . . . 8
6.2. Preference Labels . . . . . . . . . . . . . . . . . . . . 9
6.3. Text Encoding . . . . . . . . . . . . . . . . . . . . . . 9
6.4. Syntax Extensions . . . . . . . . . . . . . . . . . . . . 9
6.5. Processing Algorithm . . . . . . . . . . . . . . . . . . 10
6.6. Alternative Formats . . . . . . . . . . . . . . . . . . . 11
7. Consulting a Preference Expression . . . . . . . . . . . . . 11
7.1. Combining Preferences . . . . . . . . . . . . . . . . . . 12
8. Applicability and Legal Effect . . . . . . . . . . . . . . . 12
9. Security Considerations . . . . . . . . . . . . . . . . . . . 13
10. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 13
Keller & Thomson Expires 22 January 2026 [Page 2]
Internet-Draft AI Preference Vocabulary July 2025
11. References . . . . . . . . . . . . . . . . . . . . . . . . . 13
11.1. Normative References . . . . . . . . . . . . . . . . . . 13
11.2. Informative References . . . . . . . . . . . . . . . . . 14
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 14
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 14
1. Introduction
This document defines a common vocabulary of terms for automated
systems that process digital assets. The primary purpose of this
vocabulary is to enable machine-readable expressions of preferences
about how digital assets are used by automated processing systems in
the context of training AI models and other forms of automated
processing.
The terms defined by the vocabulary can be used to describe, in a
standardized way, the types of uses that a declaring party may wish
to explicitly restrict or allow. Preferences are then expressed as a
grant or denial of permission concerning each of the types of use
defined in the vocabulary. This ensures that preferences can be
communicated, processed, and stored in a consistent and interoperable
manner.
The vocabulary or the preferences that might be expressed do not
proscribe how automated processing systems obtain or act on
preferences. Separate documents will describe how preferences might
be associated with assets. It is designed to ensure that preference
information can be exchanged between different systems and
consistently understood.
The vocabulary is intended to be usable both where expressing
preferences results in legal obligations and where there are no
associated legal protections. That is, preferences can be expressed
to invoke specific protections, or they can be made without any
presumption of specific legal consequences. Potential legal
obligations include rights reservations made by rightholders in
jurisdictions with conditional exceptions on copyright protections.
Expressing preferences is without prejudice to applicable laws,
including the applicability of exceptions and limitations to
copyright.
2. Conventions and Definitions
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in
BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
Keller & Thomson Expires 22 January 2026 [Page 3]
Internet-Draft AI Preference Vocabulary July 2025
This document uses the following terms:
AI:
Artificial intelligence or machine learning, which are used
interchangeably in this document, refer to computer systems or
algorithms that are trained to accomplish a task.
AI Training:
Processing input data to identify statistical trends in order to
produce an AI system.
Asset:
A digital file or stream of data, usually with associated
metadata.
Declaring party:
The entity that expresses a preference with regards to an Asset.
3. Statements of Preference
The vocabulary is a set of categories, each of which is defined to
cover a class of usage for assets. Section 4 defines these
categories in more detail.
A statement of preference is made about an asset. Statements of
preferences can assign preferences to each of the categories of use
in the vocabulary. Preferences regarding each category can be
expressed either to allow or disallow the usage associated with the
category.
A statement of preferences can express preferences about some, all,
or none of the categories from the vocabulary. This can mean that no
preference is expressed for a given usage category.
Some categories describe a proper subset of the usages of other
categories. A preference that is expressed for the more general
category applies if no preference is expressed for the more specific
category.
For example, the Automated Processing category might be assigned a
preference that allows the associated usage. In the absence of any
statement of preference regarding the AI Training category, that
usage would be also be allowed, as AI Training is a subset of the
Automated Processing category. In comparison, an explicit preference
regarding AI Training might disallow that usage, while permitting
other usage within the Automated Processing category.
After processing a statement of preferences the recipient assigns
each category of use one of three preference values: "allowed",
"disallowed", or "unknown".
Keller & Thomson Expires 22 January 2026 [Page 4]
Internet-Draft AI Preference Vocabulary July 2025
In the absence of a statement of preference, all usage categories are
assigned a preference value of "unknown".
3.1. Conformance
This document -- and those documents that define concrete uses of
this vocabulary -- describe how usage preferences are associated with
assets. Conformance to the specification means following the
normative language that defines the construction and interpretation
of usage preferences. The process of obtaining preferences has very
limited scope for variation between implementations.
Variation in what usage preferences might be obtained are limited to:
* the choice of which methods for expressing preferences -- such as
those defined in [ATTACH] -- are implemented and used, and
* the terms that are in the vocabulary, as might be added by updates
to this document; see Section 6.4.
There is considerably more discretion involved in respecting
preferences. An entity MAY choose to respect these preferences when
processing assets. This is done according to both:
* an understanding of the nature of that processing and how it
corresponds to the usage categories where preferences have been
expressed, and
* the legal context that applies; see Section 8.
Usage preferences can be overridden through express agreements
between relevant parties. There are also many situations where other
priorities could override any usage preferences. For example, people
with accessibility needs might override a preference to disallow AI
Use (Section 4.4) so that they might access automated captions or
summaries. Another case might involve the use of assets for
research. Such overrides might be explicitly permitted in law, or
could be based on the judgment of individual system users.
4. Vocabulary Definition
This section defines the categories of use in the vocabulary.
Figure 1 shows the relationship between these categories:
Keller & Thomson Expires 22 January 2026 [Page 5]
Internet-Draft AI Preference Vocabulary July 2025
.-------------------------------------------------.
| |
| Automated Processing |
| |
| .-------------------------------------------. |
| | .------------------------. | |
| | | | | |
| | | | | |
| | AI Training | Generative AI Training | | |
| | | | | |
| | | | | |
| | '------------------------' | |
| '-------------------------------------------' |
| |
| .----------------. .----------------. |
| | | | | |
| | | | | |
| | AI Use | | Search | |
| | | | | |
| | | | | |
| '----------------' '----------------' |
| |
'-------------------------------------------------'
Figure 1: Relationship Between Categories of Use
4.1. Automated Processing Category
The act of using one or more assets in the context of automated
processing aimed at analyzing text and data in order to generate
information which includes but is not limited to patterns, trends and
correlations.
The use of assets for automated processing encompasses all the
subsequent categories.
4.2. AI Training Category
The act of training machine learning models or artificial
intelligence (AI).
The use of assets for AI Training is a proper subset of Automated
Processing usage.
Keller & Thomson Expires 22 January 2026 [Page 6]
Internet-Draft AI Preference Vocabulary July 2025
4.3. Generative AI Training Category
The act of training general purpose AI models that have the capacity
to generate text, images or other forms of synthetic content, or the
act of training more specialized AI models that have the purpose of
generating text, images or other forms of synthetic content.
The use of assets for Generative AI Training is a proper subset of AI
Training usage.
4.4. AI Use Category
The act of using one or more assets as input to a trained AI/ML model
as part of the operation of that model (as opposed to the training of
the model).
The use of assets for AI Use is a proper subset of Automated
Processing usage.
4.5. Search Category
Using one or more assets in a search application that directs users
to the location from which the assets were retrieved.
The purpose of defining a distinct Search category is to allow
preferences to be expressed about search applications, independent of
other categories of use. A distinct Search category allows for
preferences specific to search applications, even if the use of AI is
involved in their implementation.
The use of assets for Search is a proper subset of Automated
Processing usage.
5. Usage
The vocabulary is used by referencing the terms defined in Section 4,
directly or via mappings, in accordance with how they are defined in
this document.
5.1. More Specific Instructions
A recipient of a statement of preferences that follows this model
might receive more specific instructions in two ways:
* Extensions to the vocabulary might define more specific categories
of usage. Preferences about more specific categories override
those of any more general category.
Keller & Thomson Expires 22 January 2026 [Page 7]
Internet-Draft AI Preference Vocabulary July 2025
* Statements of preferences are general purpose, machine-readable
statements that cannot override contractual agreements or more
specific statements.
For instance, a statement of preferences might indicate that the use
of an asset is disallowed for AI Training. If arrangements, such as
legal agreements, exist that explicitly permit the use of that asset,
those arrangements likely apply, unless the terms of the arrangement
explicitly say otherwise.
5.2. Vocabulary Extensions
Systems referencing the vocabulary MUST NOT introduce additional
categories that include existing categories defined in the vocabulary
or otherwise include additional hierarchical relationships.
6. Exemplary Serialization Format
This section defines an exemplary serialization format for
preferences. The format describes how the abstract model could be
turned into Unicode text or sequence of bytes.
The format relies on the Dictionary type defined in Section 3.2 of
[FIELDS]. The dictionary keys correspond to usage categories and the
dictionary values correspond to explicit preferences, which can be
either y or n; see Section 6.2.
For example, the following expresses a preference to allow AI
training (Section 4.2), disallow generative AI training
(Section 4.3), and and expresses no preference for other categories
other than subsets of these categories:
train-ai=y, train-genai=n
6.1. Usage Category Labels
Each usage category in the vocabulary (Section 4) is mapped to a
short textual label. Table 1 tabulates this mapping.
Keller & Thomson Expires 22 January 2026 [Page 8]
Internet-Draft AI Preference Vocabulary July 2025
+========================+=============+=============+
| Category | Label | Reference |
+========================+=============+=============+
| Automated Processing | all | Section 4.1 |
+------------------------+-------------+-------------+
| AI Training | train-ai | Section 4.2 |
+------------------------+-------------+-------------+
| Generative AI Training | train-genai | Section 4.3 |
+------------------------+-------------+-------------+
| AI Use | ai-use | Section 4.4 |
+------------------------+-------------+-------------+
| Search | search | Section 4.5 |
+------------------------+-------------+-------------+
Table 1: Mappings for Categories
Any mapping for a new usage category can only use lowercase latin
characters (a-z), digits (0-9), "_", "-", ".", or "*". These are
encoded using the mappings in [ASCII].
6.2. Preference Labels
The abstract model used has two options for preferences associated
with each category: allow and disallow. These are mapped to single
byte Tokens (Section 3.3.4 of [FIELDS]) of y and n, respectively.
6.3. Text Encoding
Structured Fields [FIELDS] describes a byte-level encoding of
information, not a text encoding. This makes this format suitable
for inclusion in any protocol or format that carries bytes.
Some formats are defined in terms of strings rather than bytes.
These formats might need to decode the bytes of this format to obtain
a string. As the syntax is limited to ASCII [ASCII], an ASCII
decoder or UTF-8 decoder [UTF8] can be used. This results in the
strings that this document uses.
Processing (see Section 6.5) requires a sequence of bytes, so any
format that uses strings needs to encode strings first. Again, this
process can use ASCII or UTF-8.
6.4. Syntax Extensions
There are two ways by which this syntax might be extended: the
addition of new labels and the addition of parameters.
Keller & Thomson Expires 22 January 2026 [Page 9]
Internet-Draft AI Preference Vocabulary July 2025
New labels might be defined to correspond to new usage categories.
Section 5.2 addresses the considerations for defining new categories.
New labels might also be defined for other types of extension that do
not assign a preference to a usage category. In either case, when
processing a parsed Dictionary to obtain preferences, any unknown
labels MUST be ignored.
The Dictionary syntax (Section 3.2 of [FIELDS]) can associate
parameters with each key-value pair. This document does not define
any semantics for any parameters that might be included. When
processing a parsed Dictionary to obtain preferences, any unknown
parameters MUST be ignored.
In either case, new extensions need to be defined in an RFC that
updates this document.
6.5. Processing Algorithm
To process a series of bytes to recover the expressed preferences,
those bytes are parsed into a Dictionary (Section 4.2.2 of [FIELDS]),
then preferences are assigned to each usage category in the
vocabulary.
The parsing algorithm for a Dictionary produces a keyed collection of
values, each with a possibly-empty set of parameters. The parsing
process guarantees that each key has at most one value and
parameters.
To obtain preferences for each of the categories in the vocabulary,
iterate through the categories. For the label that corresponds to
that category (see Table 1), obtain the corresponding value from the
collection, disregarding any parameters. A preference is assigned as
follows:
* If the value is a Token with a value of y, the associated
preference is to allow that category of use.
* If the value is a Token with a value of n, the associated
preference is to disallow that category of use.
* Otherwise, a preference is not expressed for that category of use.
Note that this last alternative includes the key being absent from
the collection, values that are not Tokens, and Token values that are
other than y or n. All of these are not errors, they only result in
no preference being inferred.
Keller & Thomson Expires 22 January 2026 [Page 10]
Internet-Draft AI Preference Vocabulary July 2025
An important note about this process and format is that, if the same
key appears multiple times, only the last value is taken. This means
that duplicating the same key could result in unexpected outcomes.
For example, the following expresses no preferences:
train-ai=y, train-ai="n", train-genai=n, train-genai, all=n, all=()
If the parsing of the Dictionary fails, no preferences are expressed.
This includes where keys include uppercase characters, as this format
is case sensitive (more correctly, it operates on bytes, not
strings).
This process produces an abstract data structure that assigns a
preference to each usage category as described in Section 3.
6.6. Alternative Formats
This format is only an exemplary way to represent preferences. The
model described in Section 3, can be used without this serialization.
Any alternative format needs to define the mapping both from that
format to the model used in this document and from the model to the
alternative format. This includes any potential for extensions
(Section 6.4).
The mapping between the model and the alternative format does not
need to be complete, it only needs to be clear and unambiguous.
For example, an alternative format might only provide the ability to
convey preferences for a subset of the categories of use. A mapping
might then define that no preference is associated with other
categories.
7. Consulting a Preference Expression
After processing a preference expression (Section 6.5), an
application can request the status of a specific usage category.
A single preference expression can be evaluated for a usage category
as follows:
1. If the expression contains an explicit preference (either to
allow or disallow), that is the result.
2. Otherwise, if the usage category is a proper subset of another
usage category, recursively apply this process to that category
and use the result of that process.
Keller & Thomson Expires 22 January 2026 [Page 11]
Internet-Draft AI Preference Vocabulary July 2025
3. Otherwise, no preference is expressed.
This process results in three potential answers: allow, disallow, and
no preference. Applications can use the answer to guide their
behavior.
One approach for dealing with an "unknown" or "no preference" answer
is to assign a default. This document takes no position on what
default might be chosen as that will depend on policy constraints
beyond the scope of this specification.
7.1. Combining Preferences
The application might have multiple preference expressions, obtained
using different methods.
If multiple preference expressions are active, all preference
expressions are consulted (Section 7). This might result in
conflicting answers.
Absent some other means of resolving conflicts, the following process
applies to each usage category:
* If any preference expression indicates that the usage is
disallowed, the result is that the usage is disallowed.
* Otherwise, if any preference preference allows the usage, the
result is that the usage is allowed.
* Otherwise, no preference is expressed.
This process ensures that the most restrictive preference applies.
8. Applicability and Legal Effect
This document provides a set of definitions for different categories
of use, plus a system for associating simple preferences to each
(allow, disallow, or no preference; see Section 3).
The categories of use that are defined as part of the vocabulary are
not always clearly applicable or inapplicable to a particular system
or application. The universe of possible systems is far more complex
than any simple vocabulary is capable of describing. That means that
some discretion could be involved in deciding whether a preference
applies.
Keller & Thomson Expires 22 January 2026 [Page 12]
Internet-Draft AI Preference Vocabulary July 2025
The expression of preferences might activate regulatory or legal
consequences, which has implications for entities that consume those
preferences. Their interpretation of the meaning of different terms
could have legal ramifications. Different jurisdictions could reach
subtly different conclusions about the applicability of each category
of use to specific applications.
It is the responsibility of those that process affected assets to
understand the legal implications of their use of digital assets.
This includes understanding:
* obligations regarding how preferences are obtained (in particular,
which methods of associating preferences with content are expected
to be understood),
* the specific uses to which assets are put,
* how preferences apply to the those uses, and
* how relevant jurisdictions might interpret those preferences.
These considerations will depend on jurisdiction and the details of
the system.
9. Security Considerations
TODO Security
10. IANA Considerations
This document has no IANA actions.
11. References
11.1. Normative References
[ASCII] Cerf, V., "ASCII format for network interchange", STD 80,
RFC 20, DOI 10.17487/RFC0020, October 1969,
<https://www.rfc-editor.org/rfc/rfc20>.
[FIELDS] Nottingham, M. and P. Kamp, "Structured Field Values for
HTTP", RFC 9651, DOI 10.17487/RFC9651, September 2024,
<https://www.rfc-editor.org/rfc/rfc9651>.
Keller & Thomson Expires 22 January 2026 [Page 13]
Internet-Draft AI Preference Vocabulary July 2025
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/rfc/rfc2119>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/rfc/rfc8174>.
11.2. Informative References
[ATTACH] Illyes, G. and M. Thomson, Ed., "A Vocabulary For
Expressing AI Usage Preferences", Work in Progress,
Internet-Draft, draft-ietf-aipref-attach-02, 21 July 2025,
<https://datatracker.ietf.org/doc/html/draft-ietf-aipref-
attach-02>.
[UTF8] Yergeau, F., "UTF-8, a transformation format of ISO
10646", STD 63, RFC 3629, DOI 10.17487/RFC3629, November
2003, <https://www.rfc-editor.org/rfc/rfc3629>.
Acknowledgments
The following individuals made significant contributions to this
document:
* Cullen Miller
* Laurent Le Meur
* Leonard Rosenthol
* Sebastian Posth
* Timid Robot Zehta
Authors' Addresses
Paul Keller
Open Future
Email: paul@openfuture.eu
Martin Thomson (editor)
Mozilla
Email: mt@lowentropy.net
Keller & Thomson Expires 22 January 2026 [Page 14]