AI companies offer up to $2 million for other companies' internal emails and files

Data intermediaries are offering companies between USD 350,000 and USD 1.8 million for copies of emails, documents, and internal messages to train Artificial Intelligence. Offers for call transcripts are reportedly at USD 2 million. Proposals came from data sellers, who suggested that AI labs spend over USD 45 billion on private data. Some entities, such as Anthropic, license professional documents so labs can test model handling of knowledgeable work, partly because a large part of their business involves graduate student work. Companies like Mercor, Micro1, and Handshake lead this sector. Micro1's CEO estimated that current annual payments for data between labs, companies, and robotics firms range from USD 40 billion to USD 50 billion. The market is projected to exceed a trillion dollars annually in two to four years, with selling work data potentially becoming a more common revenue stream than advertising. A specific transaction involving Spirit Airlines' digital assets faced a dispute over the winning offer between Google and Micro1 due to employee data concerns regarding privacy protection.

Data intermediaries are offering companies between USD 350,000 and USD 1.8 million for copies of their emails, documents, and internal messages to train AI. An offer of USD 2 million was mentioned if call transcripts were included. Steven Levy reported that some claimed Anthropic and OpenAI pay media outlets these sums for old work records. Handshake offered USD 1,000 to an independent journalist for a single article, explaining professional documents are licensed for model testing and graduate student work. Mercor, Micro1, and Handshake lead the business. Ali Ansari estimated that current annual payments for data among labs, companies, and robotics firms range from USD 40 billion to USD 50 billion. Spirit Airlines' digital assets included 80,000 email accounts, 500 million internal messages, 20.6 million Microsoft SharePoint files, and one million shift records. Google offered USD 10 million for the data; Micro1 outbid it by USD 2.5 million in a dispute.

The situation reveals a significant tension between the accelerating commercial demand for proprietary data to fuel AI development and existing frameworks for data ownership, privacy, and labor law. The movement from traditional advertising revenue to selling granular work data suggests a fundamental restructuring of value in the digital economy, shifting focus from attention economics to informational asset economics. The dispute surrounding Spirit Airlines' assets highlights the friction between open market speculation and established legal/labor concerns regarding employee versus customer data protection. Furthermore, the acknowledgment that unstructured data like personal communications contains sensitive information—such as medical or family issues—undermines the simple transactional view of data; it introduces a layer where the value extracted from datasets is inseparable from individual human context and potential harm. The projection of a trillion-dollar market signals that regulatory lag is likely to be severe, forcing adaptation through evolving legal interpretations rather than static rules. What is the ultimate cost associated with the 'messiness' of corporate workflow when it is commodified for training systems?