AI’s next data race is moving inside the workplace.
WIRED reported on Nov. 6 that data brokers had approached media companies and other organizations with offers to buy or license old workplace materials, including documents, emails, Slack messages, tickets and call transcripts. One pitch claimed AI labs were paying US$350,000 to US$1.8 million for work records, while another raised the possible payment to US$2 million.
The reporting does not prove that every named AI lab is directly buying those archives. Anthropic told WIRED it does not buy workflow data from live companies like WIRED, while OpenAI did not provide a response before publication.
The target material is broader than formal documents. Mercor’s public data page lists Slack, Microsoft Teams, Gmail, Outlook, Salesforce, GitHub, Drive, Notion and Confluence as possible sources, along with workplace messages, repositories, issues, reports and workspace content. Micro1’s data partnership page says it is seeking standard operating procedures, process documentation, internal knowledge bases, CRM records, support tickets and business process artifacts from established companies. Handshake’s data partnership page similarly says documentation, workflows and project history can help AI systems learn from real work.
The amounts being advertised vary by buyer and dataset. Micro1 says payments typically range from $100,000 to more than $3 million after acceptance. Handshake’s public page lists potential business data partnerships from $100,000 to $4 million or more. Those are marketing claims by the data firms, not independently verified market prices.
The value is in the workflow, not just the file
Public text can teach general language and facts. An AI agent meant to operate inside a business needs a different kind of example: the path from request to context, decision, tool use and final deliverable. Internal records can capture false starts, exceptions, approvals and messy handoffs that polished public documents usually omit.
OpenAI’s GDPval benchmark provides context for that demand. OpenAI described GDPval as an evaluation built around real-world, economically valuable work across 44 occupations, using tasks designed to resemble professional deliverables. That benchmark is not the same as buying a company archive, but it points to the same technical appetite for realistic work tasks rather than generic web text.
In January, WIRED separately reported that OpenAI asked third-party contractors to upload real assignments and deliverables from current or previous workplaces to evaluate AI agents. According to that report, the instructions told contractors to remove personal information, proprietary or confidential data and material nonpublic information before uploading files. OpenAI and Handshake AI declined to comment for that January report.
The Spirit Airlines fight made the issue concrete
The abstract market became visible in the bankruptcy case of Spirit Airlines. WIRED reported in August that Google won the auction with a US$10 million bid to buy about 34 years of Spirit data, subject to court approval, and that the Association of Flight Attendants-CWA objected over employee privacy concerns.
The archive described in court filings, according to WIRED, included more than 1 million time-card records, more than 175,000 employee records, nearly 150,000 employee tax forms, employment contracts and litigation files, 80,000 email accounts, 17 million individually owned Microsoft OneDrive items, 20.6 million shared Microsoft SharePoint files and 500 million Microsoft Teams records.
By Nov. 3, WOWT/InvestigateTV reported that the union and more than 100 members of Congress were seeking stronger privacy protections in the proposed Google transaction. Google said, according to the report, that it was not buying personal information from Spirit and that information would either be excluded or de-identified by an independent third party before Google received it. WIRED’s Nov. 6 report said the sale was on hold after the union objection and a competing bid dispute. A Nov. 4 bankruptcy court filing moved the sale hearing to Nov. 17.
De-identification is not the whole debate
Data firms emphasize de-identification, scoping and security controls. Those safeguards may reduce risk, but they do not answer every question raised by workplace archives. Emails, Teams messages, personnel files and shared documents can contain medical details, family issues, disciplinary matters, payroll records, grievances and other information that was never created with AI training in mind.
The technical challenge is that the parts that make these datasets valuable are often the same parts that make them sensitive. AI developers want relationships between messages, files, decisions and outcomes. Workers and unions are worried that preserving those relationships can make people identifiable even when names or direct identifiers are removed.
Consent is a separate problem. Employers may control many records in company systems, but the records were often written by employees, contractors, customers and partners. If the new market treats those archives as assets, workplace data policy is likely to become a labour, privacy and governance issue rather than an IT housekeeping task.
What it means for Canadian businesses
For Canadian organizations, the news is not only about one U.S. bankruptcy dispute. The Office of the Privacy Commissioner of Canada says PIPEDA sets rules for private-sector organizations that collect, use or disclose personal information in commercial activities, and that it also applies to employee personal information at federally regulated businesses.
The same OPC page notes that Alberta, British Columbia and Quebec have substantially similar private-sector privacy laws, and that PIPEDA can still apply where personal information crosses provincial or national borders in commercial activity.
The OPC’s AI guidance also says organizations developing or using generative AI should establish legal authority for collecting and using personal information, be transparent about use and privacy risks, limit sharing of personal, sensitive or confidential information, and build privacy protections into AI systems. Those principles do not decide every possible data licensing question, but they show why workplace archives cannot be treated as ordinary spare files once AI training enters the picture.
I can suggest related Tech Help Canada articles based on the topic you’re reading now.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
