Frontenac Labs

Articles · October 22, 2026

Can your company license its business records to AI companies?

If your company has been operating for ten or fifteen years, it holds an archive that nobody outside it has seen: project files, proposals, support threads, CRM notes, drawings, reports. Some AI data programs now look for private business records like these. Here is why, which companies are likely to fit, and what to settle before you agree to anything.

Why AI developers are looking past the public web

Large language models have been trained mostly on public web text. In its investigation of OpenAI, the Office of the Privacy Commissioner of Canada found that more than 99% of the information OpenAI used to pre-train its models was crawled from publicly accessible sources, and the rest came from licensed third-party datasets [1].

That public supply is finite. Researchers at Epoch AI estimate the usable stock of public human-written text at roughly 300 trillion tokens. They project that, if current trends continue, models will be trained on datasets about that size somewhere between 2026 and 2032 [2]. The authors call this a projection, not a certainty. They note that synthetic data and better data efficiency may stretch the supply [2]. Still, it explains why developers are looking for text that isn't on the open web.

Licensing deals for content are already a market. In 2024, Reuters reported that Reddit agreed to make its content available to Google for AI training under a contract worth about US$60 million per year, according to one source [3]. Reddit is a public platform with a huge archive, so that figure says nothing about what a private company's records would be worth. It does show that AI companies pay for data they can't simply collect.

Why private business records are different

Public text tends to be opinion, news and marketing. Business records show how work actually gets done: how an engineer answers a request for information, how a deal team structures a data room, how a support agent resolves a billing dispute. Most of that never reaches the public web, which is why it can't be scraped.

Which companies are likely to fit

Buyers set their own criteria and they change, so treat this as a starting point rather than a rule. The criteria our free Data Check tests against are:

Our Data Check flags these as likely to be excluded or to need special handling: consumer personal data, health records, privileged legal files, payment card data, children's data, and classified or export-controlled material. Meeting the criteria doesn't mean a buyer will accept your data or make an offer.

Questions to settle before you say yes

1. Do you have the right to license it? Many records involve clients, partners or employees. Check client contracts, NDAs and IP clauses before anything else. For records that hold personal information, Canada's privacy regulator says organisations sourcing data from third parties should make sure that data was collected lawfully and that the third party has authority to disclose it [4].

2. What personal information is in it, and can it be removed? The Privacy Commissioner's principles for generative AI call for limiting collection to what is necessary and using anonymised or de-identified data where possible [4]. The same principles stress that information being publicly accessible doesn't put it outside privacy law [4]. In its OpenAI findings, the Commissioner found the collection from public and licensed sources overbroad [1]. Expect a careful buyer to ask how your data will be filtered, and ask them the same question.

3. Who in your company must agree? Licensing touches legal, IT, leadership and often client-facing teams. A person with real authority should approve before any application goes in.

4. What exactly leaves the building, and when? Find out what a buyer reviews first, whether files are shared at that stage, how data will be transferred and stored, and what happens if you decline the offer.

5. Who is paid, and by whom? If an intermediary introduces you to a buyer, ask how that intermediary is paid. It affects whose interests they represent.

Where to start

Begin with an inventory: which systems hold your records, how many years they go back, and which datasets are unusual for your industry. Then talk to legal about client contracts and personal information before you talk to any buyer.

Our free Data Check takes about two minutes. It gives a rough, illustrative estimate of how many records you hold and how you match the criteria above. No files are shared. It isn't an offer or a valuation, and some companies aren't accepted.

Frontenac Labs is an independent referrer and does not act for any buyer. If your company licenses data to a buyer we introduce, that buyer may pay us a referral fee. You pay us nothing. We cannot promise any price, payout or acceptance; the buyer decides. This article is general information, not legal or privacy advice.

Could your company's records be licensed?Our free Data Check takes about two minutes. No files are shared, and it isn't an offer.
Check your company's data

Sources

  1. Office of the Privacy Commissioner of Canada, PIPEDA Findings #2026-002: Overview of the Joint Investigation of OpenAI OpCo, LLC. https://www.priv.gc.ca/en/opc-actions-and-decisions/investigations/investigations-into-businesses/2026/pipeda-2026-002-overview/
  2. Villalobos et al., Epoch AI, Will we run out of data? Limits of LLM scaling based on human-generated data (ICML 2024). https://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-data and https://arxiv.org/abs/2211.04325
  3. Reuters, "Exclusive: Reddit in AI content licensing deal with Google," 22 Feb 2024. https://www.reuters.com/technology/reddit-ai-content-licensing-deal-with-google-sources-say-2024-02-22/
  4. Office of the Privacy Commissioner of Canada, Principles for responsible, trustworthy and privacy-protective generative AI technologies. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/