Why AI buyers want training data from defunct and closing companies

Why a closed company's records interest AI buyers

AI labs and data buyers want records from closed and closing companies for the same reason they want them from any established business: years of real work captured in tickets, code reviews, project files, approvals and email threads, with the outcome of each piece of work known. A closed company's archive adds one thing. The story is complete, from the first customer to the last invoice.

The demand comes from a change in what AI systems are asked to do. Models that answer questions learned mostly from public text. Agents that carry out multi-step work need examples of how that work is actually done inside companies, and that material is thin on the public web. Researchers at Epoch AI estimated the effective stock of public human-written text at roughly 300 trillion tokens and projected that language models could fully use it between 2026 and 2032 if current trends continue. It is a forecast with wide uncertainty, but it explains why permissioned, non-public records are treated as a scarce input.

Older archives already have a licensing precedent outside business records. In July 2023 the Associated Press agreed to license part of its text archive dating back to 1985 to an AI developer, with financial terms undisclosed. A company's operational archive is a different asset, but the idea that well-kept historical records can be licensed is not new.

Training, fine-tuning and evaluation: how licensed records can be used

The three uses below are distinct, and one dataset can serve more than one of them.

Quality matters as much as volume. The US Copyright Office's report on generative AI training, released as a pre-publication version in May 2025, notes that model performance depends heavily on the quality of training data and considers how practical licensing approaches are. Records of real work, kept in structured business systems with outcomes attached, differ sharply from scraped web text.

Which records from a closing company matter most

Breadth helps. Strong companies typically keep records across 10-15+ systems, and the links between them, such as a ticket tied to a customer tied to an invoice, are often worth more than any single system. Project-based businesses add their own layer, which construction contractor insolvency covers for contractors.

What a closure destroys, and in what order

Records rarely disappear in one decision. They erode as the company winds down:

1. Staff leave. The administrators who know where data lives and how to export it are often among the first to go.
2. Subscriptions lapse. SaaS accounts are cancelled to save cash, and what the vendor keeps afterwards, and for how long, depends on its own terms.
3. Devices are wiped. Laptops go back to lessors or are sold, taking local files with them.
4. Domains and email expire. When the domain lapses, access to the mail system tied to it goes too.
5. Storage bills go unpaid. Cloud archives disappear when the account closes.

Each step is sensible on its own. Together they can leave nothing to license. That is why the window to act is before the cancellation list is executed, not after the last day of operations. Advisers who see a closure coming can use the scripts in how to talk to a client about closing their business.

What happens to the data after licensing

Licensing is not a sale. The company, or whoever now controls its assets, keeps ownership and grants a license that is typically exclusive for AI training for an agreed term. De-identification and redaction requirements are agreed before any work begins, and nothing is delivered until the agreement is executed and the company authorizes it. The company receives one all-in price, with SourceX's fee included, as a one-time payment, typically within about 60 days of invoicing once the buyer selects the data.

When a company has entered a formal process, the person who authorizes that license is the fiduciary. How to choose an ABC assignee covers the records questions to settle before an assignment, and bankruptcy data sales in the AI era covers the court-side issues.

Limits: when a closed company's records cannot be licensed

Many defunct companies hold records that should never be licensed, and buyers will not take them.

• Promises to customers. FTC staff have written that companies' promises not to use customer data for undisclosed purposes, such as training models, are enforceable, whether made in privacy policies, terms of service or elsewhere. That post is staff guidance, not a rule, but the principle is clear: check what the company promised before any customer-facing data is included.
• Client-owned material. Agencies and outsourcers often hold records that belong to their clients.
• Sensitive personal data. Archives made up mainly of consumer personal data, or of patient health information without authorization or de-identification, do not fit.
• Size and history. The baseline still applies after a closure: 50+ full-time employees at peak (contractors excluded) and several years of documented operations.
• Nothing to export. Deleted archives, or systems nobody can still reach, end the conversation.
• Already licensed or generated. Data already licensed for AI training, and records created with AI in order to sell them, are red flags.

This is general information, not legal, tax or financial advice. Confirm with your own counsel before relying on it for a specific company.

What this means for referral partners

The partners best placed here hear about a closure before the cancellation list is final: outside accountants, fractional CFOs, turnaround advisers, lenders, investors and IT providers. The task is to ask one question in time, not to judge the data. A partner's role ends at the introduction, with no exports, no uploads and no descriptions of what the records contain. If a trustee is already in place, overlooked intangible assets in chapter 7 gives the trustee's view.

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company, paid only after the buyer pays and SourceX receives its fee. No reward is guaranteed.

Next step

Check a closing company against the who qualifies baseline or run it through the company fit checker, then register as a partner and make the introduction while the systems still exist.