Data ReserveAI data licensing research
Section 01 / Direct questions

Frequently asked questions

Data Reserve research deskSection 01Page 13 of 16Last updated 4 November 2026

Does owning the website mean we own all its content?

No: copyright ownership is distinct from possession, and contributions can have separate owners under US law. Review the underlying contracts and components rather than treating the domain as a rights certificate. US Copyright Office — Title 17, Chapter 2: Copyright Ownership and Transfer

Can a retrieval licence include training automatically?

The answer depends on the actual agreement, including definitions and incorporated terms. For negotiation, request a use-by-use schedule because retrieval and model training have different technical functions. Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Is a marketplace listing enough evidence to fund full preparation?

No; require a task, recipient and acceptance requirement before committing the full budget. AWS documentation describes distribution and offer mechanics, which do not themselves demonstrate a purchase for your inventory. AWS — Product subscriptions in AWS Data Exchange

Is there a standard AI archive price per article?

No defensible universal per-article benchmark was established in this research. Compare real proposals on identical scope and use the sourced market-data document to understand the limits of disclosed deal evidence.

Does an NDA prevent a buyer from training on our sample?

An NDA should not be assumed to resolve permitted AI use; inspect the actual purpose and reuse clauses with counsel. Pair the written restrictions with screened, limited access and a named close-out procedure, consistent with NIST data-governance principles. NIST AI 600-1 — Generative Artificial Intelligence Profile, July 2024

Can we promise a model will forget our content at expiry?

Do not make an unconditional promise without model-specific technical evidence. Machine unlearning concerns removing training influence, while access revocation and file deletion are different operations. NIST CSRC — Machine unlearning

Can the rights issue be solved by removing names?

Removing names is not a copyright clearance process and does not answer every privacy or identity question. The FTC Everalbum order is a historical example of remedies reaching models developed from affected user material, not a universal rule for every archive. FTC — Everalbum: final decision and order, 7 May 2021

What should we do first if a partner wants exclusivity this week?

Ask for the exact scope, fee base, performance obligations, conflict disclosures and release conditions, then identify the evidence missing from the appointment decision. Use a bounded extension or narrower proposed mandate if proper review cannot be completed; the choice must reflect your own legal and commercial circumstances.

Sources cited on this page

  1. US Copyright Office — Title 17, Chapter 2: Copyright Ownership and Transfer ↗copyright.gov
  2. Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks ↗arxiv.org
  3. AWS — Product subscriptions in AWS Data Exchange ↗docs.aws.amazon.com
  4. NIST AI 600-1 — Generative Artificial Intelligence Profile, July 2024 ↗nvlpubs.nist.gov
  5. NIST CSRC — Machine unlearning ↗csrc.nist.gov
  6. FTC — Everalbum: final decision and order, 7 May 2021 ↗ftc.gov