Your Old Slack Threads Are Now a Prized Asset — AI Labs Are Turning Startup Archives Into Training Gold

Somewhere on a server you’ve long forgotten about, the Slack threads you wrote at a startup years ago — the bug triage debates, the customer escalations, the “how do I get this API to work” messages — may now be worth real money. According to a report from The Information published August 13, the data demand from OpenAI and Anthropic has turned startups’ old Slack threads, support tickets, and internal emails into suddenly prized assets, because this kind of enterprise operational data can help AI agents learn how software is actually used in the real world.

It is the strangelest kind of asset boom: companies discovering that their operational exhaust — the unglamorous byproduct of just doing work — has become a commodity that frontier AI labs want to buy.

Why AI labs want your chat logs

The reason is structural. The first generation of large language models was trained on the public internet: web pages, books, code repositories. That corpus is largely tapped, and it has a critical blind spot for the current phase of AI — agents. An agent that is supposed to operate software on your behalf needs to understand not what software looks like in documentation, but how people actually use it: the messy sequences of clicks, commands, workarounds, error reports, and negotiations that happen inside Slack threads, Jira tickets, and email chains.

That data doesn’t exist on the open web. It lives inside companies. And as Forbes reported in April, defunct startups are now being effectively liquidated for their Slack archives, Jira tickets, and email threads — operational exhaust that AI labs treat as valuable training material. The Information’s August reporting confirms the demand has only intensified and broadened: it’s no longer only the archives of dead companies changing hands, but a widening market where old Slack threads, tickets, email, and even video messages from operating businesses are “suddenly in high demand.”

The market that emerged from startup graveyards

The mechanics of this market came into public view in April 2026, when Fast Company reported that SimpleClosure — a startup-shutdown service — had launched a tool allowing companies to sell their data wealth, from Slack archives to email chains, to AI labs. Gizmodo reported at the time that defunct startups were “pawning off” their former employees’ conversations for payouts of up to $100,000.

Read that number again: a company with no product, no customers, and no IP left to sell could still walk away with a six-figure check — provided its employees had been sufficiently chatty on the clock. Futurism crystallized the uncomfortable core of the story: the rush to build workplace AI agents has spawned an entire industry for repurposing employee data from dead startups, and the people who wrote those messages have no claim, no cut, and often no idea.

What’s changed by August

Three things make this week’s reporting more than a rerun of the April story.

First, the buyers are named. The Information’s reporting centers on demand from OpenAI and Anthropic — the two labs whose enterprise-agent ambitions are the most aggressive. When the most capitalized AI companies in the world signal they will pay for a data category, a market forms fast.

Second, the seller profile has widened. April’s market was a liquidation story: dead companies monetizing the last thing they owned. August’s reporting describes startups finding that their operational records are “suddenly” valuable — which changes the incentive structure for companies that are still alive. Your archive is no longer something to purge when you migrate tools; it’s inventory.

Third, the data types have expanded. Text was the beachhead. The demand now reportedly extends to video messages and ticket histories — anything that captures how work actually flows through software. Every new modality an agent must master becomes a new category of raw material someone must sell.

The consent problem nobody solved

The legal and ethical footing here is thinner than the market’s momentum. Employees who wrote those Slack messages typically never consented to their resale for AI training. Companies generally own workplace communications under employment agreements — but “we own it” and “our employees expected us to sell it to train third-party AI models” are very different claims. Privacy advocates flagged this exact gap in April, and nothing reported since suggests it has been closed.

There’s also a competitive dimension: a startup that sells its operational data is potentially handing a distilled record of its internal playbook — how it supported customers, how it shipped, where it struggled — to the same labs building the agents that may compete with the next startup. The $100,000 payout buys a lot of runway, but it’s worth asking what it sells.

Why this matters beyond the nostalgia of old chats

If the agent era is real, operational data is to agents what web text was to chatbots: the substrate. That reframes several things businesses routinely do. Deleting old workspaces to save on SaaS fees now has a hidden opportunity cost. Data-retention policies are quietly becoming data-asset policies. And the due-diligence checklist for acquisitions may soon include a line item for the training value of the target’s collaboration archives.

For individuals, the takeaway is starker. The most detailed record of how you actually work — your problem-solving patterns written out in your own words across years of internal chat — can be sold, without your knowledge, to train systems that may one day replicate parts of your job. The old consolation was that this only happened to failed companies. That consolation is gone.

What to watch

Expect three pressure points. Pricing transparency: as more sellers enter, expect data brokers and shutdown services to publish rate cards for archive types. Regulation: the EU AI Act’s transparency regime and data-governance provisions are already forcing provenance questions across the AI supply chain, and resold employee communications sit squarely in that path. And consent frameworks: the first platform to build real revenue-sharing for the humans behind the data — rather than just the companies that hosted it — will have found the moral high ground of an otherwise gritty market.

Until then, assume your old Slack threads are being read by someone. Just not your former colleagues.