AIG-049 Copyright Policy and Rights Reservation Compliance
Description
A copyright policy covering every general-purpose AI model placed on the market exists in a single document with a named owner, and each of its measures is implemented. Web crawling for training data reproduces only lawfully accessible content, does not circumvent technological protection measures or subscription and paywall restrictions, excludes sites recorded as persistent infringers, and reads and honours machine-readable rights reservations expressed through the Robot Exclusion Protocol and the other protocols the policy names, with the crawler's identity and the reservations it honours published. Technical safeguards limit the model's reproduction of protected training content in its output, the acceptable use terms or the documentation accompanying the model prohibit infringing use, and a published contact point and complaints mechanism for rightsholders exist, with each complaint recorded and answered within a defined period.
Rationale
The policy is one document because the Code of Practice asks for one and because an auditor otherwise has to assemble it from crawler configuration, a terms page and a support queue. The crawl-side measures are testable from the crawler itself: its user agent, the robots.txt handling and the exclusion list are configuration, and a rightsholder's reservation can be checked against a crawl log. The output-side safeguard is tested by prompting for known protected content. Where crawling is outsourced the same measures bind the contractor and the contract carries them. Boundary with GOV-015: that control protects the organisation's own use of licensed assets; this one governs collection from and reproduction of other people's. Boundary with AIG-013: that control records each dataset's licence status; this one is the conduct that decides what enters the datasets. gpai-provider seat (ADR-031).
Applicability (9 profiles)
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
A general-purpose model providers duty (gpai-model-provider, ADR-046). The deployer takes the model documentation and the published training summary the provider issues into its AIG-032 assessment.
Art.53(1)(c): the copyright policy and rights reservation compliance, for every general-purpose model, with no open-source relief.
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
A general-purpose model providers duty (gpai-model-provider, ADR-046). The deployer takes the model documentation and the published training summary the provider issues into its AIG-032 assessment.
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
The copyright policy and crawler conduct of a general-purpose model provider are the gpai-provider seat (ADR-046). A SaaS providers own licensed assets and training data licences are GOV-015 and AIG-013.
Framework Mappings (8)
| EU-AI-Art.53.3 | GPAI Model Obligations — Copyright Compliance Policy | full |
| COP-C-1 | Copyright policy | full |
| COP-C-1.1 | Draw up, keep up-to-date and implement a copyright policy | full |
| COP-C-1.2 | Reproduce and extract only lawfully accessible copyright-protected content when crawling the World Wide Web | full |
| COP-C-1.3 | Identify and comply with rights reservations when crawling the World Wide Web | full |
| COP-C-1.4 | Mitigate the risk of copyright-infringing outputs | full |
| COP-C-1.5 | Designate a point of contact and enable the lodging of complaints | full |
| MP-4.1-006 | AI Technology and Legal Risk Mapping | MP-4.1-006 | informative |
Evidence (3)
The copyright policy for the organisation's general-purpose models, in a single document with a named owner, stating the crawl-side, output-side and complaints measures.
Example: General-Purpose Model Copyright Policy v1.2, owner the Head of Data Acquisition, approved 30 June 2026.
Test: Verify: (1) the policy is one document naming an owner and covering every model placed on the market, (2) it states the lawful-access rule, the non-circumvention rule and the exclusion of persistently infringing sites, (3) it names the reservation protocols honoured, with the Robot Exclusion Protocol among them, and where the crawler identity and honoured reservations are published, (4) it states the output-side safeguard and where infringing use is prohibited, (5) it states the rightsholder contact point, the complaints mechanism and the response period, (6) where crawling is contracted out, the contract carries the same measures.
Crawler configuration and crawl logs showing the published user agent, robots.txt handling, the exclusion list and a reservation honoured in practice.
Example: Crawler fleet configuration export of 2 September 2026 and the crawl log for the week of 24 August 2026.
Test: Verify: (1) the crawler identifies itself with the published user agent, (2) robots.txt directives addressed to that agent are fetched and applied, shown by a disallowed path absent from the crawl log, (3) the exclusion list is loaded and a listed site is absent from the log, (4) paywalled or access-restricted content is not fetched, shown by the absence of authenticated sessions in the crawler configuration, (5) a reservation expressed through another protocol the policy names is honoured in the same way.
The rightsholder complaints register with each complaint, its substantiation, the action taken and the response date against the defined period.
Example: Rightsholder complaints register 2026-H1, 14 complaints, exported 3 July 2026.
Test: Verify: (1) the contact point is published where the model is offered and reaches the register, (2) each complaint carries the date received, the substantiation check, the action taken and the date answered, (3) each response falls within the defined period or carries a recorded reason, (4) a complaint found substantiated shows the reservation, exclusion or safeguard change that followed, (5) a manifestly unfounded or duplicate complaint carries that determination and its reason.
Questions (3)
Does a single copyright policy with a named owner cover every general-purpose model you place on the market?
A policy spread across crawler documentation, terms of use and a support procedure is a no until it is brought into one document. Answer for the models as such, not for products built on them.
Which of the following measures are implemented?
Options follow the order of the policy measures from collection to complaint. Where crawling is outsourced, answer for what the contractor is bound to and shown to do.
How are rightsholder complaints handled?
Options run from the fullest arrangement to none. A mechanism that exists but has never been used is the first option only if a register exists to receive complaints and a period is defined.