Your old Slack history now has a market price, Google is deciding which AI features schools may refuse, and Microsoft thinks you can stop reading agent-written code. Some of these claims are better supported than others.
1. Your old Slack history is now inventory
Cold pitches have started arriving in WIRED’s inbox. One claims labs like Anthropic and OpenAI pay media companies $350K to $1.8M for old documents, emails, Slack messages and tickets. Anthropic told WIRED it doesn’t buy workflow data from live companies. OpenAI hadn’t responded by publication.
The clearest look at the market is the Spirit Airlines bankruptcy sale: 80,000 email accounts, 500 million internal messages and 20.6 million SharePoint files. Google bid $10 million, Micro1 topped it by $2.5 million, and the sale is now on hold after the flight attendants’ union said it included their private information.
The brokers say de-identification solves privacy. EPIC’s counsel replies that unstructured messages, like ones about doctor’s appointments, are hard to scrub. That dispute is the real test. My read is that bankruptcies and startup wind-downs are the easiest supply, because the people who wrote the messages have the least leverage. Founders should check who controls their data if the company is ever sold off.
Treat the market size with caution. The $40–50 billion annual figure comes from Micro1’s CEO, and WIRED couldn’t verify it.
2. Google lets schools refuse Gemini, but not AI search
Ars Technica’s headline contains the whole tension: schools can keep Search while skipping Gemini, but AI search is non-negotiable. We only have that headline and a one-line summary, which calls it more choice, but “not the choices anyone wants.“
The distinction matters. Declining a standalone assistant is easy. Declining an AI layer built into the tool schools already use for research is not. My inference is that the opt-out is real but sits on the less important product.
The actual terms, and what schools lose by objecting, aren’t in the excerpt. Until they are, treat this as a warning about platform leverage, not a documented ultimatum.
3. DigiCert says customer data sat exposed for 16 months
The Register reports that DigiCert has admitted a 16-month security lapse exposed customer information. The number is the story. Sixteen months is longer than a year, which suggests the problem went unnoticed through a lot of ordinary operations.
The text we have for this report is mostly navigation links, not the article. We can’t say what information was exposed, how the lapse happened, who found it, or whether anyone exploited it. Any account of a botched migration or a specific cause would be speculation.
Practically, DigiCert customers should look for the company’s notice and see which categories of their data it covers. Whether the lapse reached anything that affects certificate issuance itself is the question to ask first.
4. Fox closed its $22B Roku deal months ahead of schedule
When Fox announced the Roku deal in June, it projected a close in the first half of 2027. It finished on November 6. Two months ago the Justice Department issued a second request for more documents, and reports this month said approval was near. TechCrunch’s account doesn’t say how the review ended.
Fox’s stated payoff is plain: access to Roku’s more than 100 million households, better ad targeting, and less dependence on traditional TV distribution. It says the combined company is now the third-largest US television business by viewership. Roku founder Anthony Wood says the mission doesn’t change.
The incentives do. A neutral-looking platform now belongs to a company that produces programming and wants viewer data for its own ads. My view is that developers and programmers who rely on Roku should watch whether placement, data access or terms ever favor Fox content. Nothing in the reporting says that has happened.
5. Microsoft’s PDD Kit moves code review into the spec
Microsoft Research’s pitch is that agents write too much code to read line by line, so humans should review intent and let a verifier check the code. PDD Kit interviews the developer, turns the answers into a Verus specification for Rust, proves the agent’s code meets it, and falls back to tests where proof can’t reach.
In the demo, a developer gave a two-sentence request about login in a roughly 5,000-line Rust store app. The tool asked whether the right password must always work, not just whether a wrong one must fail, a gap an agent might have settled silently. The developer never read the generated code line by line.
That is real progress, but the guarantee is narrower than the pitch. The proof covers only what the spec says. Lookups, bcrypt, token signing and the clock were trusted, and six assumptions needed sign-off. So the bottleneck moves from reading code to writing the right spec and accepting the right assumptions. That is a better review task, but not a free one. This is also Microsoft’s own demo of one slice of one app.