Inferensys

Glossary

TDM Opt-Out

A machine-readable protocol enabling content owners to declare that their copyrighted works are reserved for Text and Data Mining, overriding general crawling permissions to prevent unauthorized AI training ingestion.
Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.
TEXT AND DATA MINING RESERVATION

What is TDM Opt-Out?

A machine-readable protocol enabling content owners to declare that their copyrighted works are reserved for Text and Data Mining, overriding general crawling permissions to prevent unauthorized AI training ingestion.

A TDM Opt-Out is a technical declaration that reserves the right to perform text and data mining on a copyrighted work, explicitly prohibiting its use for commercial AI model training. It functions by overriding general crawling permissions granted in robots.txt through a specific machine-readable signal, such as the TDM-Reservation: 1 HTTP response header or a dedicated tdm-opt-out directive, thereby communicating to compliant automated agents that ingestion is forbidden without a separate license.

This protocol addresses the legal gap between standard web crawling and AI training ingestion under the European Union's CDSM Directive, which requires rights holders to expressly reserve their rights. Unlike a broad Disallow rule that merely blocks access, a TDM Opt-Out establishes a legally significant preference signal that, when paired with a Data Processing Agreement, provides a technical foundation for enforcing copyright compliance and preventing unauthorized inclusion in permissioned corpora.

PROTOCOL MECHANICS

Key Characteristics of TDM Opt-Out

The TDM Opt-Out is a machine-readable declaration that reserves the right to prevent text and data mining, overriding general crawling permissions. Below are the core technical and legal characteristics defining this protocol.

01

Machine-Readable Reservation

The TDM Opt-Out functions as a protocol-level declaration rather than a human-readable legal notice. It is implemented via:

  • robots.txt directives: Specifically targeting AI crawler user-agents with Disallow rules.
  • HTTP response headers: Utilizing X-Robots-Tag: tdm-reservation to signal opt-out status at the page level. This structured format allows compliant automated agents to parse the reservation without manual interpretation.
02

Overrides General Crawling Permissions

A critical distinction exists between indexing and mining. A site may allow indexing via Googlebot for search visibility while explicitly disallowing AI training crawlers.

  • Granularity: The protocol supports granular path exclusion, reserving specific directories for TDM while leaving others accessible.
  • Precedence: The TDM reservation takes logical precedence over broad Allow rules for specific AI user-agents, creating a specific exception to general access.
03

Legal Basis in EU Copyright Law

Article 4 of the EU Directive on Copyright in the Digital Single Market (CDSM) provides the statutory framework. Key legal mechanics include:

  • Article 4(3): Explicitly grants rights holders the ability to reserve their rights to prevent text and data mining.
  • Opt-Out Mechanism: The reservation must be expressed in a machine-readable format, making the technical protocol a direct extension of legal compliance.
  • Commercial Restriction: This right applies specifically to commercial entities and for-profit AI training, distinguishing it from scientific research exceptions.
04

Compliance Verification

The protocol relies on automated adherence by crawlers. Verification involves:

  • Server-side logging: Monitoring access logs for 200 OK responses to disallowed paths from known AI user-agents.
  • User-agent string analysis: Cross-referencing traffic against a maintained user-agent blocklist to identify non-compliant bots.
  • Provenance checks: Auditing downstream training datasets for the presence of content from domains with active TDM reservations.
05

Relationship to Preference Signals

The TDM Opt-Out operates alongside broader digital consent mechanisms:

  • Global Privacy Control (GPC): While GPC signals a general objection to data sales, the TDM Opt-Out is a content-specific directive targeting the act of mining itself.
  • Consent Management Platforms (CMPs): These can syndicate TDM preferences, but the technical enforcement still relies on the server-side protocol.
  • Do Not Scrape: The TDM Opt-Out is the legally grounded, machine-readable evolution of the conceptual 'Do Not Scrape' signal.
06

Enforcement and Limitations

The protocol's primary limitation is its reliance on voluntary compliance by crawler operators. Enforcement challenges include:

  • Non-compliant actors: Malicious scrapers ignoring robots.txt entirely.
  • Attribution difficulty: Proving that a specific model trained on opted-out content without a verified provenance chain.
  • Jurisdictional scope: The CDSM directive applies to EU member states, creating a fragmented global enforcement landscape for non-EU AI developers.
TDM OPT-OUT PROTOCOL

Frequently Asked Questions

Clarifying the technical and legal mechanisms that allow content owners to reserve their rights against unauthorized text and data mining for AI training.

A TDM Opt-Out is a machine-readable protocol enabling content owners to declare that their copyrighted works are reserved for Text and Data Mining, overriding general crawling permissions to prevent unauthorized AI training ingestion. Technically, it is implemented via the robots.txt parser or HTTP response headers to communicate a reservation of rights to automated agents. Specifically, the TDM Reservation Protocol allows rights holders to signal that, while a page may be indexed for search, it cannot be used for commercial AI training or computational analysis. This is distinct from a simple Disallow directive, which blocks all crawling; a TDM Opt-Out permits indexing but restricts the purpose of the crawl, aligning with the legal framework of Article 4 of the EU's DSM Directive.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.