# What your CPA should check before you license data

> What public deals paid, and the checks your CPA or fractional CFO can run on your records and the risks before you license anything to an AI lab.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/partners/cpas-and-cfos/ai-data-licensing-guide  
Published: 2026-10-05  
Updated: 2026-10-14

Most advice about AI covers tools and policy. This guide covers your company's own records, what AI labs pay to use them, and the checks your CPA or fractional CFO can run before you decide anything.

## What AI data licensing is

AI data licensing means your company lets an AI lab use its records for a set time, under written rules. The records are things like Slack threads, email, CRM notes and support tickets. Your company keeps ownership.

> **Licensing means a lab uses your records for a while, under rules.** Rent the room. Keep the house.

Labs borrow real work to teach software agents how work gets done. Epoch AI, a research group, [reports](https://epoch.ai/gradient-updates/state-of-rl-envs) that labs spend heavily on simulated workplaces for agents.

It also describes lab partnerships with product companies and does not discuss real company archives. So demand is real but still being tested.

> **Labs want real work to learn from.** Borrowed. Returned. No late fees.

## What do public deals show?

Public prices are few. We track every one we can find, and your CPA can check each source.

- **Small shutdowns.** SimpleClosure sold archives from nearly 100 small companies at $10,000 to $100,000 per company, according to [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916).
- **A long history.** Cielo24 sold 13 years of Slack, Jira and email for "hundreds of thousands" of dollars, per the same report.
- **An airline archive.** Google won the auction for Spirit Airlines' archive with a $10 million bid, and court approval is pending, per [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/). Mercor bid $7.5 million, and micro1 later offered $12.5 million, per [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/).
- **Stated program tiers.** micro1 lists $100,000 and up for qualified companies, $500,000 and up for large ones, and $1 million and up for highly unique data, per [its page](https://www.micro1.ai/data-partnerships). That is micro1's own claim.
- **Six figures.** Troveo says typical full-company deals start at six figures, per [its page](https://www.troveo.ai/business-data). That is also self-reported.

**$10K to $100K** What small shutdowns have sold their archives for ([source](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916))

  

**$10M** Google's winning bid for one airline's archive. ([source](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/))

No public price list exists, and [Troveo says so itself](https://www.troveo.ai/resources/sell-data-to-ai-companies). A range with a source is more useful to you and your CPA than a single number.

Figure: scale, Reported archive prices: A small shutdown's archive $10K to $100K, Google's winning bid $10M (emphasized). Google's winning bid, still pending court approval, is a hundred times the top of a small shutdown's archive. Source: Gizmodo, Time and Fortune.

_Public figure, see source ([source](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916))_

_The tall bar is an airline. Your company is probably the short one._

## What risks should your CPA flag?

Ask your CPA or CFO to walk through these with you, in plain words.

- **Other people's information.** Email and chat hold customer and vendor secrets, often under NDAs. Daniel Goldberg, a partner at Frankfurt Kurnit, also flags trade secrets, waived privilege and protected employee talk in a [2026 note](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes).
- **Privacy law.** Under California rules, licensing personal data can count as a sale unless names and details are stripped out, per [Freshfields](https://www.freshfields.com/en/our-thinking/blogs/a-fresh-take/california-expands-its-data-broker-rules-in-the-b2c-context-unpacking-california-102jpiw).
- **Vendor terms.** Salesforce limited bulk export and AI use of Slack data pulled through its tools in 2025, per [Hunton](https://www.hunton.com/insights/legal/salesforce-locks-down-slack-data-time-to-review-your-slack-api-terms).
- **Old promises.** The FTC says quiet changes to terms for AI training can be [unfair or deceptive](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive).

See the [risks page](/risks) for more in plain words.

## Do your records fit?

Your CPA sees most of these signs in the books already. Go through them together.

- 30 or more staff. micro1 sets that floor, per [micro1's partnership page](https://www.micro1.ai/data-partnerships).
- Years of Slack, email, CRM or ticket history that are still stored.
- A sale, merger, raise or wind-down on the horizon.
- A line in the books for SaaS tools that hold records, such as chat and ticketing.
- Someone on staff who knows where the admin logins live.

_Your SaaS expense report is a map of the attic._

Our [company data map](/tools/data-map) helps you list the systems first. Bring the one-page result to your CPA.

## Where to start

Run the free [data check](/business) first. It takes about three minutes and gives you a result and one next step. Share the result with your CPA or CFO.

If you want to talk it through, [send Matt a note](/contact). To see who licenses records and what each one asks for, read the [buyer comparison](/buyers).

Your annual review with your CPA is a good place for this, and so is your AI policy. The next guide covers [the records line](/partners/cpas-and-cfos/the-data-line-item) step by step.

_Glim counts. Glim does not file taxes._

## Questions owners ask

### Is AI data licensing money taxable income?

In most cases, yes. IRS Publication 525 says an amount included in your income is taxable unless a law specifically exempts it. How your company books and reports AI data licensing payments is your CPA's call, and Vaultfish does not give tax advice.

### Does licensing records to an AI lab change who owns them?

No. A data license rents records out under set rules, and your company keeps ownership. The license contract spells out the details, and your lawyer reads it before anyone signs.

### How big can an AI data licensing deal get?

Public AI data deals run from $10,000 for a small shutdown's archive to a $10 million winning bid for one airline's archive. Most companies land well under the top figure, and many records are worth little.

### My CPA's firm does our audit. Can my company still license its records?

Yes. An audit relationship limits your CPA's fee, and their independence lead checks any other work. A CPA whose firm audits or reviews your books takes no referral fee from Vaultfish, and your company can still license its records.

### Can I check my records for AI value without my CPA?

Yes. The Vaultfish data check is free, takes about three minutes and needs no advisor.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._