Shutting down? What your Slack, email, Jira and code are worth, and who buys them

Why AI labs want a closing company’s records

A closing startup’s Slack, email and code now have buyers. Demand for real company archives surged in 2025 as AI labs hit the limits of public and synthetic datasets, MIT Technology Review reported on October 13, 2026.

The main use is training AI agents to do office work. Labs build reinforcement learning (RL) environments, which TechCrunch describes as “training grounds that simulate what an AI agent would be doing in a real software application”. The money behind them is large: in September 2025, The Information reported, via TechCrunch, that Anthropic leaders had discussed spending more than $1 billion on RL environments over the following year. Epoch AI, which interviewed 18 people across environment startups and labs, calls enterprise workflows “a major growth area,” puts typical task costs at $200 to $2,000 per task, and cites one interviewee’s estimate of around $300,000 for a high-fidelity replica of a product like Slack.

A real archive is what makes those simulations realistic. Surge AI uses company data to build RL gyms where agents practice tasks such as debugging IT architecture or drafting appellate briefs, and its lead for this work told MIT Technology Review he looks for archives that include disagreements and tricky judgment calls. SimpleClosure, which sells shutdown assets through its AssetHub, lists its buyers as including “leading AI foundation labs, reinforcement learning environment builders”.

One caveat belongs up front. The same MIT Technology Review piece notes that real archives avoid errors common in synthetic data, but that the size of the performance gain is unclear and no company has published comparison tests. Buyers quote wildly different numbers, Surge’s lead told the magazine, because valuations rest on each buyer’s “research theses”.

What sells and what does not

SimpleClosure says buyers want assets that “reflect how real products and services get built”: the spec, the ticket, the pull request, the release, the customer email. The more of that chain you can deliver, the more useful it is.

What the buyers say they value, in their own words:

• Variety and depth over volume. Surge’s lead told MIT Technology Review that one company’s Slack history with enough variety can be worth more than another company’s Slack, email and calendars combined.
• Headcount, years and tool use. Sunset, a wind-down service, values archives on headcount, years in operation, data volume and how the team used its tools, per the same article.
• Reasoning, not just output. Turing’s Project Lazarus lists decision-making, meaning the trade-offs and reasoning behind choices, next to code, tickets and runbooks.
• Size and language thresholds. micro1 asks for at least 30 employees and data that is mostly in English; its program for closed companies asks for 50+ peak employees and Slack or Teams data. Mode sets its floor at 30+ employees for most businesses. Scale calls 40+ employees its “preferred starting point”.
• Code by volume. Code-only buyers can price mechanically. pre.dev publishes $100 per million source tokens and $40 per million git-history tokens, with a worked example of $6,000 for one repository.

MIT Technology Review reports that the firms supplying the data employ data scientists to scrub personal information and exclude HR records, and that buyers say they scrub too.

Published price ranges versus what shutdown deals close at

Two sets of numbers circulate, and they measure different things.

What buyers publish. These are ranges stated by each company on its own site, checked on October 18, 2026. They are ceilings and illustrations, not offers, and most are pitched at operating companies.

The full, sortable comparison with fees, ingestion method and sources for every cell is on the buyer index.

What shutdown deals have actually paid. Reported transactions sit mostly at the low end of the published ranges:

• SimpleClosure told Forbes it processed nearly 100 deals in the prior year, with payouts of $10,000 to $100,000 per company (as summarized by Gizmodo); Forbes’ lead example, cielo24, sold 13 years of Slack, Jira and email for “hundreds of thousands of dollars”.
• One crypto founder received about $25,000 for a codebase, with a workspace-data sale pending at an estimated $15,000.
• SimpleClosure says AssetHub has handled more than 150 deals and expects to return over $10 million to startups by the end of 2026, and that 47% of its deals now include code or data sales.
• Ooak Data says its closed deals range from $20,000 to “several hundred thousand dollars” (company figures, not limited to shutdowns).

Even the low end matters at closure. SimpleClosure’s H1 2026 Shutdown Report puts median cash at closure, among companies it closed, at $13,000 for non-AI companies and $30,000 for AI companies. The pool of sellers is also growing: estimates from Carta and AngelList show shutdowns rising 26% to 56% a year; Carta alone counted 966 US startup shutdowns in 2024, up from 769 in 2023, among its own customers.

SimpleClosure’s own CEO warns sellers that a headline figure “doesn’t tell you what this buyer will pay”. Treat published ranges as a reason to ask, not a valuation.

Who decides: board, investors, creditors, trustees

The founder rarely decides alone. Who signs depends on where the company is in its wind-down.

Dissolving a Delaware corporation itself needs a board resolution and a vote of a majority of the outstanding stock (section 275). If a data sale is on the table, put it on the same agenda so the approvals line up.

Two practical points follow. First, MIT Technology Review reports that proceeds typically go to senior creditors and investors, not employees; founders holding common stock are paid only after creditors and any preferred liquidation preference. Second, buyers check authority. Ooak Data says the signer is “whoever holds signing authority for the entity at that moment” and asks for the authority document up front.

Bankruptcy sales draw scrutiny. MIT Technology Review reported that Google outbid Mercor in Spirit Airlines’ bankruptcy with a $10 million bid for the airline’s data, including 100 million emails, 500 million Teams messages and 176,000 employee records. The Next Web reported on October 9, 2026 that the sale still needed the bankruptcy court’s approval, with a hearing set for October 14, after a consumer privacy ombudsman reviewed it and micro1 filed notice of a competing bid. Frankfurt Kurnit described the proposed sale as drawing objections from the flight attendants’ union and from a software vendor that says some of the data belongs to it.

Timing: export before you cancel anything

The most avoidable way to lose a data sale is to cancel the tools before exporting. Each platform has its own clock.

The rule is simple: keep every subscription active until you have a complete export, a signed license or a decision not to sell. A few more weeks of subscription fees cost far less than the smallest deals reported above. For the Google steps in detail, see how to cancel Google Workspace and keep your data.

Step-by-step checklist

1. Pause every cancellation. Tell finance not to cancel Google Workspace, Microsoft 365, Slack, Atlassian, GitHub or your CRM yet. Check that no auto-renewal is about to lapse.
2. Confirm who can sign. Get a board resolution that authorizes exploring a data license. Read your charter and investor agreements for consent rights. If an ABC or Chapter 7 filing is coming, talk to the assignee or trustee before you contact buyers.
3. Inventory what you have. List each system, its plan level, the years of history, peak headcount and the main language. This is what every buyer asks first.
4. Read your own promises. Pull the privacy policy and employee handbook in force when the data was collected, plus customer contracts, DPAs and NDAs. Mark anything that restricts transfer.
5. Export with native admin tools. Use Google’s Data Export, Slack’s own export (public channels on Free and Pro; Slack grants private-channel and DM exports on Business+ or Enterprise only on application, described as for legal and regulatory compliance, so state your purpose accurately), the Microsoft 365 admin tools, Atlassian backups and full git clones with history. Keep the exports encrypted and limit access to one or two people.
6. Set aside what you will not sell. HR and legal folders, customer data you processed as a vendor, and anything regulated. It is easier to exclude it now than to argue about it in diligence.
7. Get several written offers from different end buyers. Two intermediaries may resell to the same lab, so ask each one who the end buyer is. You can ask us to compare buyers for you; we never ask for data, credentials or exports.
8. Compare on terms, not headline price. Use the comparison grid under How to compare offers below, or our data licensing offer checklist.
9. Have counsel review the license and get final approval. The board approves the signed version, and proceeds go through the creditor waterfall.
10. Close out cleanly. After delivery, get written confirmation that raw data was deleted on the agreed schedule, then cancel the subscriptions.

Legal blockers to clear before you sign

This is a map of the issues, not legal advice. Frankfurt Kurnit’s October 2026 client note is a clear public summary, and most points below come from that note.

• Your own privacy statements. Companies “must comply with the privacy statements in effect when the data was collected.” Quietly adopting looser terms may be “unfair or deceptive” in the FTC’s eyes.
• California employees. The CCPA covers employees, applicants and contractors, and transferring workplace data for money is a sale unless it meets the CCPA’s de-identification definition.
• Labor law. Employee chats about pay, scheduling and working conditions may be protected concerted activity under Section 7 of the National Labor Relations Act, unionized or not.
• Other companies’ secrets. Email and chat carry partners’ and customers’ confidential information under NDAs you may not be free to license.
• Customer contracts and DPAs. If you processed data as a service provider, your customer agreements and DPAs decide whether it can be sold. Check them before anything else.
• Privilege and trade secrets. A sale can waive privilege over legal communications and weaken trade secret protection.
• De-identification limits. Scrubbing names from millions of messages is far harder than from a database, and a disciplinary thread stays sensitive without a name on it.

Platform terms decide how data should leave. Slack’s API terms bar third-party app providers from using API data “to train a large language model” and from bulk-exporting message data without an additional agreement. Google’s Workspace API User Data and Developer Policy bars using Workspace API data to “create, train, or improve a machine learning or artificial intelligence model beyond that specific user’s personalized model”, and Google’s verification guidance says this covers data that is aggregated, anonymized or derived. Ask every buyer in writing whether it ingests through your own admin export or through an API connection, and keep the answer with your records.

Regulated professions. Tax preparers who knowingly or recklessly disclose return information commit a misdemeanor punishable by a fine of up to $1,000 (up to $100,000 in identity-theft cases) and up to one year in prison under 26 U.S.C. 7216. Law firms risk privilege, and health businesses hold patient data. Data Licensing Index does not route client data from law, accounting, tax or health firms.

Staff in the EU, UK or Canada. Sweden’s data protection authority says an employer normally cannot use employee consent as a legal basis, because employees are not free to refuse (IMY, in Swedish). The EDPB’s Opinion 28/2024 addresses the processing of personal data in the context of AI models. Data Licensing Index does not route employee email or chat from staff based in the EU, UK or Canada. Code and documentation without personal data are a separate question for local counsel.

How to compare offers

Founders have little visibility into who buys their data and little assurance that offers are fair. The fix is to put every offer on the same grid.

Also ask who is paid for the introduction. Several buyers pay referral fees to whoever brings them a company; micro1, for one, publishes up to $50K per referral. That means an adviser steering you to one buyer may be paid by that buyer. Our own revenue model works the same way, so we mark every paying buyer on its row and name it to you before any introduction. See our disclosure.

How this guide was built

Every figure above links to the page it came from, checked on October 18, 2026. Ranges are published by the companies themselves, and we have not verified any company’s payout history. Our criteria and update schedule are on the methodology page. When you are ready, tell us what you have and we will compare buyers for your company. We never touch your data.