AI data market
Failed startups selling Slack and email archives to AI: what happened
By SourceX Editorial · Reviewed by Noah Loul ·
Short answer
Failed startups selling Slack and email archives to AI became public in April 2026, when Forbes described closed companies' chat, ticket and email records being sold to AI developers, often through wind-down firms, to train and test workplace agents. Concerns centered on employee privacy, customer material and signing authority. Sellers should license reviewed exports, never workspace access.
Key takeaways
- Forbes reported in April 2026 that closed startups' chat, ticket and email archives were being sold to AI developers, often through wind-down firms.
- Buyers wanted the archives because they show a whole company working, which suits training and testing workplace agents.
- Slack's API terms bar third-party apps from bulk export and LLM use, so the owner's own admin export is the realistic route.
- Direct messages, HR channels, shared customer channels and pasted credentials are the usual problem areas.
- Automated de-identification needs human sampling, because detection tools do not promise to find everything.
What happened with failed startups and their archives?#
Failed startups' leftover Slack workspaces, email archives and project tickets became a visible source of AI data in April 2026, when Forbes reported that defunct companies were selling them to AI developers. Its central example was cielo24, whose former CEO used the wind-down firm SimpleClosure to close the company; the remaining internal chat, project-tracking tickets and emails in employees' cloud drives then became items for sale.
SimpleClosure told Forbes it had processed nearly 100 deals for defunct companies in the past year, and that parts of its Asset Hub tool for selling code, chat archives and emails were still in beta while it refined how it strips personal information. A Newser summary of the report said cielo24 sold 13 years of Slack, Jira and email archives. The payment figures in the coverage came from individual deals and company statements, and they say little about what another archive would bring.
Buyers wanted the archives because they show a company working: engineers debating a fix, support staff escalating a problem, managers approving a change. Forbes reported that such records feed reinforcement learning gyms, simulated workplaces where AI agents practice tasks using real documents and messages.
How did the archive market take shape?#
The archive market took shape between late 2024 and late 2026, as platform terms closed one route to workplace data and wind-down firms opened another. The table lists the dated public events behind it and what each means for a founder or wind-down officer holding an archive.
| Date | What happened | What it means for a seller |
|---|---|---|
| December 10, 2024 | Slack updated its App Developer Policy to make explicit that using data to train an LLM is prohibited | Third-party apps are not a route to training data |
| May 29, 2025 | Slack's API terms changed to bar bulk export through the APIs, persistent archives and LLM use, as summarized by Hunton Andrews Kurth; existing apps had until June 30, 2025 | A buyer's app cannot pull the workspace; the owner's admin export is the route |
| September 16, 2025 | TechCrunch reported that leading AI labs wanted more reinforcement learning environments | Demand grew for records that show whole workflows |
| April 16, 2026 | Forbes reported closed startups' chat, ticket and email archives being sold to AI developers, including through the wind-down firm SimpleClosure | The market became public, and so did the privacy questions |
| August 17, 2026 | SiliconANGLE reported that Google agreed to buy Spirit Airlines' internal business data, including emails and Teams messages, in a bankruptcy auction | Large estates joined small startups as sellers |
| August 19, 2026 | Forbes reported that wind-down startups now offer to buy closing companies' data for AI training | Founders may receive offers before anything has been scoped |
Workspace export, API pull or admin access: what actually leaves?#
The route an archive takes out of its system decides how much control the seller keeps. A scoped export that the seller runs and reviews is the only route that shows exactly what left and when.
Slack's own documentation sets the limits. Owners and admins on every plan can export public channels as JSON, but exports that include private channels and direct messages are available only on Business+ and Enterprise plans, and owners must apply to use them. Exports contain links to files rather than the files, and free workspaces permanently lose messages older than one year. Email exports run through Google Workspace or Microsoft 365 admin tools, so check current documentation before promising a buyer any scope.
| Route | What it captures | Main risk |
|---|---|---|
| Standard export of public channels | Messages and file links from public channels, as JSON | Files need a separate collection step; free workspaces hold only the last year |
| Export including private channels and direct messages | Private conversations as well as public channels | Business+ or Enterprise only, by application, and holds the most sensitive content |
| API pull by a buyer's app | Whatever the token's permissions reach | Slack's API terms bar third-party apps from bulk export and LLM training |
| Admin credentials handed to a buyer | Everything, including settings and integrations | No control over scope and no record of what was taken |
| Mailbox exports from Google Workspace or Microsoft 365 | Full threads, attachments and calendar items | Outside parties' messages and attachments in nearly every thread |
What concerns did the coverage raise, and what else should sellers expect?#
The coverage raised employee privacy most directly: Marc Rotenberg of the Center for AI and Digital Policy told Forbes that the privacy issues in selling anonymized workplace messages are substantial, given how heavily employees rely on tools like Slack. Five related concerns follow from the way these deals are done, quickly and by people with little time left at the company. Each is easier to handle before an export than after.
- Employee privacy: direct messages, HR channels and personal disclosures in casual channels.
- Customer confidentiality: Slack Connect channels, escalations quoting customer contracts, and files customers sent.
- Authority: whether a founder of a dissolved company could sign, and whether investors or creditors had to approve.
- Vendor terms: whether workspace and API terms allowed exported content to go to a third party for AI use.
- Secrets: API keys, passwords and tokens pasted into channels and emails over the years.
- Re-identification: whether removing names was enough when messages describe people, projects and events in detail.
Controls to insist on before any archive leaves#
The controls worth insisting on before an archive leaves are the ones a seller can check afterward. Presidio, an open-source tool for detecting personal details, warns in its own documentation that automated detection gives no guarantee of finding all sensitive information, so sampling by a person belongs in every plan.
Write each control into the license or its schedules, and keep the evidence that it was carried out with the company's records.
| Control | Why it matters | How to check it |
|---|---|---|
| Channel-by-channel and mailbox-by-mailbox scope | Keeps sensitive areas out by default | Written scope schedule approved by the seller |
| Direct messages excluded unless specifically reviewed | Direct messages hold the most personal content | Export settings and a sample of delivered files |
| De-identification with human review | Automated tools miss some personal details | Sampling log and a written privacy record |
| Secret scanning | Credentials in old messages may still work | Scanner report and confirmation that keys were revoked |
| No re-identification clause | Bars attempts to match records to people | License term with remedies |
| Stated permitted use and deletion | Limits how far the archive travels | License schedule and deletion confirmation at term end |
Illustrative: a founder narrows a whole-workspace offer#
Illustrative: a fictional bid-management software startup for commercial contractors shuts down when its largest customer moves to a competitor. A data intermediary offers to buy the entire Slack workspace and asks for an admin seat to pull it. The former CEO, now acting as wind-down officer, declines the admin request and asks counsel to confirm who may sign for the company.
The final package is narrower than the offer. It covers engineering and product channels, Jira issues and the shared support inbox, with names replaced by role tokens and contractor client names replaced by placeholders. Private channels, direct messages, the sales team's mailboxes and every shared customer channel stay out. A secret scan finds old cloud keys in an engineering channel, which are confirmed revoked and redacted.
Outcome: a non-exclusive license with a deletion term, a written privacy record, and a package the founder can explain plainly to former colleagues who ask.
How SourceX approaches Slack and email archives#
SourceX treats chat and email as their own record family, reviewed apart from tickets or code because the privacy questions differ. A founder or wind-down officer starts by describing the workspace, its channel names, mailbox types and date ranges, without sending a single message.
From there the SourceX five-step transaction applies: Supply, Rights, Preparation, Approval and Delivery, with the company or estate approving each stage. Records are licensed rather than sold outright, and the SourceX Evidence Packet attached to the package shows a buyer what was included, what was removed and who authorized release.
Frequently asked questions
Did the startups sell their archives outright?
Coverage described the archives as sold, and contract terms were rarely public. The difference matters: an outright sale passes control to the buyer, while a license keeps ownership with the company or estate and can limit use, sharing and duration. A license also lets the same records support more than one deal.
Are direct messages ever included?
Rarely, and only after specific review. Direct messages hold the most personal material in a workspace, from health and family news to complaints about colleagues. Careful packages exclude them by default and draw instead from channels where work was discussed openly.
Do former employees have to consent?
It depends on the company's policies and notices, the employment and privacy laws where employees worked, and what remains after preparation. Some laws may apply even after a company closes. Counsel should assess this deal by deal; removing personal details and excluding private conversations reduces the issue but may not settle it.
What if the workspace has already been deleted?
Then the archive is probably gone. Some companies find partial copies in eDiscovery exports, legal hold collections or backups, but these need the same rights review as the original. Value survives only where records were preserved before subscriptions ended.
Does an operating company face the same questions?
Yes, with more room to answer them. An operating company still has the people who know which channels are sensitive, admins who can run scoped exports and leaders who can approve. It can decide scope calmly rather than under the time pressure of a shutdown.
Sources
- Forbes reported on April 16, 2026 that cielo24's former CEO used SimpleClosure to close the company, after which its internal chat, project-tracking tickets and emails became items for sale to AI developers; SimpleClosure said it had processed nearly 100 deals for defunct companies in the past year; parts of its Asset Hub tool were in beta while it perfected PII stripping; the records feed reinforcement learning gyms; and Marc Rotenberg said the privacy issues are substantial. Source
- Newser, summarizing the Forbes report, said cielo24 sold 13 years of Slack, Jira and email archives. Source
- In a changelog entry dated December 10, 2024, Slack said it was updating its App Developer Policy, making explicit that the use of data to train an LLM is prohibited. Source
- According to Hunton Andrews Kurth, the Slack API Terms of Service were modified effective May 29, 2025 to prohibit bulk exporting of data accessible through Slack's APIs, persistent copies or archives, and use of such data in large language models; the terms applied from June 30, 2025 to existing apps. Source
- Slack's API Terms state that a provider of an application offered outside its own organization may not use API Data to train a large language model and may not bulk export Slack message and file data except where an additional agreement allows it. Source
- TechCrunch reported on September 16, 2025 that leading AI labs are demanding more reinforcement-learning environments. Source
- SiliconANGLE reported on August 17, 2026 that Google agreed to buy Spirit Airlines' internal business data in a bankruptcy auction. Source
- Court records cited in reporting describe the Spirit Airlines data as emails, Microsoft Teams messages, OneDrive files and SharePoint items. Source
- Forbes reported on August 19, 2026 that startups now offer to buy wind-down companies' data for AI training. Source
- Slack's export help article says owners and admins on all plans can export public channel messages and file links in JSON; exports including private channels and direct messages are available only on Business+ and Enterprise and require an application; exports include links to files rather than the files. Source
- Slack says messages and files more than one year old are permanently deleted from free workspaces. Source
- Presidio's documentation warns that because it uses automated detection mechanisms, there is no guarantee it will find all sensitive information, and additional systems and protections should be employed. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.