Steven Levy's latest Backchannel newsletter for Wired opens with a cold email. A sender named Nick told him that Anthropic and OpenAI were paying media companies like Wired a one-time fee of $350,000 to $1.8 million for copies of their old work records, including documents, emails, Slack messages and tickets. A second sender, Cora, wanted call transcripts too and floated $2 million. She also claimed public data is "exhausted" and that labs are spending more than $45 billion on private data.

Levy went looking for the business behind those emails and found it was much bigger than he'd guessed. Read the full piece at Wired.

We read it with some recognition. Outlever, the company that publishes State of AI, recently got a pitch of its own. We're sharing it below.

A crowded field of buyers

Levy names a lot of companies, and most of them are on the buying side.

Handshake offered a freelance writer $1,000 for her reporting notes on a single story. She turned it down. A Handshake representative confirmed the offer to Levy and said the company licenses professional documents from many fields, journalism included, so labs can test and train models on real knowledge work. Handshake says a large share of its business is buying graduate students' papers and coursework.

Sunset works with startups that are shutting down. Its pitch to one of Levy's founder friends said it scrubs and anonymizes the data, then brokers the deal with the labs.

Mercor, founded in 2023, started out as an AI recruiting company. CEO Brendan Foody told Levy the market is about to grow fast and that Mercor removes personally identifiable information in a way that can be verified. Micro1 CEO Ali Ansari went further. He put annual spending on data by labs, enterprises and robotics companies at $40 billion to $50 billion, and predicted it will pass $1 trillion a year within two to four years. Levy couldn't verify the first number. Ansari also claimed about half of Y Combinator startups are data companies. Levy counted 40 of the 65 companies in the Fall 2026 batch.

Then there's Tenkara, a two-year-old company that reached Levy through a cold pitch. Its head of commercialization, Jonah Stillman, met him in San Francisco and confirmed that calls like these are a big part of his job. Tenkara wants the data to improve an AI model that runs office operations for manufacturers. Levy admits he has no idea how Wired's archives would help with that.

The labs keep some distance

The pitches drop Anthropic and OpenAI by name. Anthropic told Levy there is a big market for data, but that it doesn't buy workflow data from live companies like Wired. OpenAI didn't respond before publication.

That statement is narrower than it sounds. It doesn't cover data from companies that have already shut down, and it doesn't say where a dataset goes after a broker buys it.

Spirit Airlines shows what's at stake

The biggest deal in Levy's piece came out of a bankruptcy. Spirit Airlines put its digital records up for sale: 80,000 email accounts, 500 million internal messages, 20.6 million SharePoint files and a million time-card records. Google reportedly bid $10 million. Micro1 came in $2.5 million higher, and the two ended up disputing who won. The sale is now on hold after the flight attendants' union said the records contain its members' private information.

Mayu Tobin-Miyaji, general counsel at the Electronic Privacy Information Center, told Levy that EPIC filed an amicus brief backing the union. Her argument is that anonymization doesn't fix the problem, because emails and Teams messages are full of personal details about health and family that don't sit in tidy fields where software can find them. Venture capitalist Gordon Ritter, who has taken up the issue, said the industry has no norms yet and is moving faster than anyone can write them.

The pitch in our inbox

Here is the email Outlever received. We've blacked out the sender's name and email address.

Hi,

We're looking to license de-identified business data from Outlever (Slack threads, support tickets, Jira issues and product docs) for AI research. Indicative estimates for companies like yours run from about $150K to over $1M, depending on headcount and years of records.

You keep owning your data, and personal information is removed before anything is shared.

Open to a quick chat? Just reply here or email me at ████████████████████████████. If it's not relevant, let me know and we won't reach out again.

██████ ████████ Renaissance Data Lab

The details line up closely with what Levy describes. The list of wanted material (Slack threads, support tickets, Jira issues, product docs) is the same operating record his sources are chasing. The price range runs a bit lower than the one Wired was quoted, and the email says the number scales with headcount and years of records. It frames the deal as a license, says Outlever would keep ownership and promises personal information gets stripped out first.

The email went to company leadership. Nobody asked the people who wrote those Slack threads and closed those tickets, which is Levy's point too. The workers whose messages get sold are usually the last to hear about it.

Our take

If you run a company with a few years of Slack history, expect one of these emails. Before you reply, get answers to a few questions.

  • Who is the actual buyer? Pitches name Anthropic and OpenAI, and Anthropic says it doesn't buy from live companies.

  • What does "de-identified" mean in practice, and who checks it? EPIC's warning about unstructured messages applies to your Slack as much as Spirit's Teams.

  • What do your employee agreements and customer contracts say? Support tickets carry your customers' information along with your own.

Then there's the question of what the data is for. These records train AI agents to do the work that produced them. Mercor's motto is "Organizing human intelligence to power the AI economy." The people whose work fills those archives usually aren't asked, and they usually aren't paid.