Classification and Tagging System for Digital Content to Preserve Clean Datasets for Machine Learning
draft-williams-ai-content-tagging-00

Network Working Group                                      K. Williams
Internet-Draft                                  Independent Researcher
Intended status: Standards Track                         June 22, 2025
Expires: December 24, 2025

     Classification and Tagging System for Digital Content to 
              Preserve Clean Datasets for Machine Learning
                 draft-williams-ai-content-tagging-00

Abstract

   This document specifies a classification and tagging system designed
   to identify and preserve the provenance of digital content (text,
   audio, video, and other media) to ensure the integrity of training
   datasets for machine learning systems.  The framework described
   herein aims to support a standardized mechanism for tagging data
   with metadata that specifies whether the content was human-generated
   or AI-generated.  This enables the exclusion of AI-generated data
   from training corpora where human-originated material is required,
   and it facilitates the maintenance of clean, verifiable sources for
   future AI development.

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/.

   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 December 24, 2025.

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.

Williams                Expires December 24, 2025               [Page 1]

Internet-Draft       AI Content Classification System         June 2025

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   3
   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   4
   3.  Metadata Structure  . . . . . . . . . . . . . . . . . . . . .   4
   4.  Implementation Mechanisms . . . . . . . . . . . . . . . . . .   5
   5.  Implementation Examples . . . . . . . . . . . . . . . . . . .   5
   6.  Trust and Verification  . . . . . . . . . . . . . . . . . . .   8
   7.  Application to Machine Learning Pipelines  . . . . . . . . .   8
   8.  Security Considerations . . . . . . . . . . . . . . . . . . .   9
   9.  Privacy Considerations  . . . . . . . . . . . . . . . . . . .   9
   10. IANA Considerations . . . . . . . . . . . . . . . . . . . . .  10
   11. References  . . . . . . . . . . . . . . . . . . . . . . . . .  11
       11.1.  Normative References . . . . . . . . . . . . . . . . .  11
       11.2.  Informative References . . . . . . . . . . . . . . . .  11
   Author's Address  . . . . . . . . . . . . . . . . . . . . . . . .  12

1.  Introduction

   With the proliferation of generative AI models producing vast
   amounts of synthetic content, it is increasingly difficult to ensure
   the quality and originality of training datasets for future AI
   systems.  This phenomenon, commonly referred to as "model collapse"
   or "data poisoning," occurs when models are trained on outputs of
   other models, compounding errors and losing alignment with human-
   authored knowledge and intent.

   As Rear Admiral Grace Hopper stated in her 1982 NSA address, every
   data record must include an identifier.  In keeping with this
   foundational principle, this RFC proposes a metadata-based
   classification system that can be attached to digital content at the
   time of its creation or publication to ensure traceability,
   discoverability, and reliability.

2.  Terminology

   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.

   HGC (Human-Generated Content):  Data or media created entirely by
      human effort, without the use of AI assistance.

   AIGC (AI-Generated Content):  Content generated in part or in full
      by artificial intelligence systems.

   Metadata Tag:  An identifier attached to content that specifies
      generation origin, timestamp, authorship, and optional
      cryptographic signature.

Williams                Expires December 24, 2025               [Page 2]

Internet-Draft       AI Content Classification System         June 2025

   Clean Dataset:  A dataset composed exclusively of HGC, validated for
      provenance and integrity.

3.  Metadata Structure

   Each piece of digital content MUST be accompanied by a metadata
   record conforming to the following schema:

   version:  The version of this metadata specification.

   origin:  Identifies the method of content generation.  Valid values
      are "human", "ai", or "hybrid".  "Hybrid" indicates partial AI
      involvement.

   author:  The creator or generating entity of the content.

   creation_timestamp:  ISO 8601 formatted timestamp of content
      creation.

   license:  The licensing terms under which the content is
      distributed.

   checksum:  SHA-256 hash of the content for integrity verification.

   signature:  Optional cryptographic signature for validation.
      RECOMMENDED for human content to validate integrity via digital
      signature.

   toolchain:  Optional field indicating the tools or AI models used
      in content generation.

   model_identifier:  Optional field specifying the specific AI model
      used for AIGC.

4.  Implementation Mechanisms

   The metadata MAY be embedded using one or more of the following
   methods:

   *  HTTP headers (e.g., X-Content-Metadata)

   *  Sidecar XML files for downloadable assets

   *  Embedded tags within HTML meta elements

   *  ID3v2 tags for MP3/MP4 audio files

   *  Exif/XMP tags for image and video files

Williams                Expires December 24, 2025               [Page 3]

Internet-Draft       AI Content Classification System         June 2025

5.  Implementation Examples

5.1.  HTTP Header for Human-Generated HTML Page

   The following example shows the HTTP header format for a human-
   generated HTML page:

   GET /article.html HTTP/1.1
   Host: example.com
   ...
   X-Content-Metadata: <metadata><version>1.0</version><origin>human</o
   rigin><author>Jane Doe</author><creation_timestamp>2025-06-21T08:00
   :00Z</creation_timestamp><license>CC-BY-4.0</license><checksum>3d2e
   61...</checksum><signature>BASE64_PGP_SIGNATURE</signature></metada
   ta>

5.2.  Sidecar File for Image

   For an image file named "sunrise.jpg", the accompanying sidecar
   metadata file would be "sunrise.jpg.meta.xml":

   <metadata>
     <version>1.0</version>
     <origin>ai</origin>
     <author>AI Art Generator</author>
     <creation_timestamp>2025-06-20T19:23:00Z</creation_timestamp>
     <checksum>abc456...</checksum>
     <toolchain>StableDiffusion-v3</toolchain>
     <model_identifier>sd-v3.1</model_identifier>
   </metadata>

5.3.  HTML Meta Tags

   The following example shows HTML meta tag implementation:

   <meta name="X-Content-Origin" content="human">
   <meta name="X-Content-Author" content="John Smith">
   <meta name="X-Content-Timestamp" content="2025-06-22T12:34:00Z">
   <meta name="X-Content-Signature" content="BASE64_SIGNATURE">

5.4.  Audio File with ID3 Tags

   For audio files, the following ID3v2 tags SHOULD be used:

   *  TXXX:Content-Origin = "human"

   *  TXXX:Content-Author = "Podcast Host"

   *  TXXX:Toolchain = (not present for human-generated content)

   *  TXXX:Checksum = "8f3a2b..."

Williams                Expires December 24, 2025               [Page 4]

Internet-Draft       AI Content Classification System         June 2025

6.  Trust and Verification

   Verification SHOULD be achieved through the following mechanisms:

   *  Public key infrastructure (PKI) for digital signature validation

   *  Trust registries for known human content sources

   *  Canonical checksums published to public hash registries

   Content consumers SHOULD verify signatures and checksums when
   available to ensure the integrity and authenticity of metadata
   claims.

7.  Application to Machine Learning Pipelines

   Dataset construction pipelines MUST implement the following
   requirements:

   *  Filter and exclude content with origin values of "ai" or "hybrid"
      where HGC-only datasets are mandated

   *  Retain metadata lineage for dataset auditing and transparency

   *  Support backtracking of samples to their original signed content

   *  Maintain logs of all filtering decisions for compliance and
      auditing purposes

8.  Security Considerations

   Several security considerations apply to this specification:

   *  Malicious actors may attempt to falsify metadata.  Digital
      signatures, cryptographic checksums, and origin auditing are
      necessary to prevent tampering.

   *  Repositories SHOULD maintain audit logs of metadata ingestion and
      content origin verification.

   *  The integrity of the metadata itself must be protected through
      cryptographic means where authenticity is critical.

   *  Trust anchors and certificate authorities used for signature
      verification must be carefully managed and regularly audited.

Williams                Expires December 24, 2025               [Page 5]

Internet-Draft       AI Content Classification System         June 2025

9.  Privacy Considerations

   Privacy protections must be considered in the implementation of this
   system:

   *  Personally identifiable information (PII) SHOULD NOT be embedded
      in metadata.

   *  Where authorship must be asserted, anonymized or pseudonymous
      cryptographic identities SHOULD be used.

   *  Content creators must have control over what identifying
      information is included in metadata tags.

10.  IANA Considerations

   This document requests the following IANA actions:

10.1.  New MIME Type Registration

   This document requests the creation of a new metadata MIME type:

   Type name:  application
   Subtype name:  x-content-metadata+xml
   Required parameters:  None
   Optional parameters:  charset
   Encoding considerations:  UTF-8 encoding is RECOMMENDED
   Security considerations:  See Section 8 of this document
   Interoperability considerations:  None known
   Published specification:  This document
   Applications that use this media type:  Content management systems,
      digital archives, machine learning platforms
   Additional information:  None
   Person and email address to contact for further information:
      Keenan Williams <keenanwilliams@gmail.com>
   Intended usage:  COMMON
   Restrictions on usage:  None
   Author:  Keenan Williams
   Change controller:  IETF

10.2.  HTTP Header Field Registration

   This document requests the registration of the following HTTP header
   fields:

   *  X-Content-Origin

   *  X-Content-Signature

   *  X-Content-Toolchain

Williams                Expires December 24, 2025               [Page 6]

Internet-Draft       AI Content Classification System         June 2025

   These header fields are intended for use in identifying content
   provenance as specified in this document.

11.  References

11.1.  Normative References

   [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/info/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/info/rfc8174>.

11.2.  Informative References

   [HOPPER]   Hopper, G., "Future Possibilities: Data, Hardware,
              Software, and People", NSA Lecture Series, 1982,
              <https://www.youtube.com/watch?v=si9iqF5uTFk>.

   [ISO8601]  International Organization for Standardization, "Date and
              time -- Representations for information interchange -- Part
              1: Basic rules", ISO 8601-1:2019, February 2019.

   [NIST-FIPS-180-4]
              National Institute of Standards and Technology, "Secure
              Hash Standard (SHS)", FIPS PUB 180-4,
              DOI 10.6028/NIST.FIPS.180-4, August 2015,
              <https://doi.org/10.6028/NIST.FIPS.180-4>.

   [RFC3986]  Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform
              Resource Identifier (URI): Generic Syntax", STD 66,
              RFC 3986, DOI 10.17487/RFC3986, January 2005,
              <https://www.rfc-editor.org/info/rfc3986>.

Author's Address

   Keenan Williams
   Independent Researcher

   Email: keenanwilliams@gmail.com