- Firms that close down startups now sell their code and workplace messages to AI labs, according to MIT Technology Review and SimpleClosure.
- The buyers want records of real work. No company has yet published tests showing that this data beats the alternatives.
- Leaders should ask what their own workplace tools hold, and who could sell it if the company fails.
When a company fails, its servers go dark. Its records may not. A growing group of firms now sells what a dead startup leaves behind to AI labs, and the buyers want something specific: how people actually worked.
What happened
SimpleClosure helps startups shut down. Its Asset Hub platform, described in the company's launch post, lets founders license source code to vetted buyers. A second offering, workspace data, was still in beta in that post. It covers documents, communications and institutional knowledge.
MIT Technology Review reports that SimpleClosure added data sales in April. It says Asset Hub has handled over 150 deals. The buyers range from large AI labs to firms that build practice workplaces for AI agents. Established tech companies training models of their own are also in the mix.
One founder, Keelan Miskell, closed a crypto startup this year. He told the magazine that his code sold for about $25,000. A further payment for his workspace data from Google, Notion, Figma and Slack is still pending, likely near $15,000. He had not expected the data to be worth much.
Why a dead company's chat history has value
MIT Technology Review reports that AI labs have hit the limits of public and synthetic data. So they are buying real workplaces. Surge AI, for example, builds what the magazine calls "reinforcement-learning gyms." These are simulated workplaces where a client's AI agent practices tasks, such as debugging a complex IT system or drafting a legal brief.
A simulation needs realistic material. Nick Heiner, who leads Surge AI's work on these environments, looks for archives with disagreements and tricky judgment calls. He says one company's Slack history can be worth more than another's Slack, email and calendars combined, if it shows variety and depth. The report adds that real archives avoid errors common in synthetic worlds, such as dates that do not exist.
Sunset, a rival firm, prices a client's assets by head count, years in operation, data volume and how the team used its tools. Its CEO, Brendan Mahony, calls this "operational exhaust data."
The idea: your daily work is the raw material
Most people write a Slack message to get something done. They do not write it as a lesson for a machine. This market treats those messages as exactly that.
Think of a salvage yard after a building is demolished. Doors, fittings and wiring that once served the occupants are sold on to new users. For companies, the leftovers are messages, drafts and arguments. They outlive the firm that made them.
The value is not yet proven. MIT Technology Review reports that no company has released tests comparing real-world data training with the alternatives. Heiner says offers vary widely. Each buyer is betting on its own theory of how many usable training tasks a given amount of code, such as 100,000 lines, will yield.
Bigger deals, and the legal doubts
Larger sales are happening anyway. In August, Google beat Mercor, another AI data company, in a bidding contest tied to Spirit Airlines' bankruptcy. The archive spanned decades. Half a billion Teams messages sat in it, alongside 100 million emails and 176,000 records about employees.
Bloomberg reported in September that SpaceX's AI unit is weighing purchases of data from struggling startups, MIT Technology Review says.
The legal footing is uncertain. The union representing Spirit's flight attendants opposed the deal, citing privacy. Greg Ewing, a lawyer who counsels AI startups, uses the phrase "legally radioactive" for this kind of data. MIT Technology Review says suppliers hire data scientists to strip out personal details and leave out HR files. In the part of the report we reviewed, there is no evidence on how well this works.
What leaders should ask
Employees wrote these messages. In the part of the report we reviewed, there is no indication whether staff are told or asked before an archive is sold. It does show that Spirit's flight attendants' union objected on privacy grounds. That is a question for leaders to answer, not a finding. The report describes a market that is growing, not a breach.
Put these questions to your legal, security and IT teams:
First, what do our Slack, email, Notion and Figma accounts hold, and for how long? Data you delete on schedule is less likely to be sold later.
Second, who controls those accounts if the company is sold or wound down? Check what contracts, privacy notices and employee agreements say about transfer, and whether staff would be told.
Third, if we buy AI tools or models, can the vendor tell us where its training data came from?
A company's records now have a market. Decide what they should be worth, and to whom, before someone else does.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: SimpleClosure.
