AI companies’ new race: Now ready to spend billions on data of human skills and everyday tasks

Washington (Staff Reporter) — AI companies have already scraped much of the public internet. Now they are pursuing something even more valuable: data showing how humans actually work. For this purpose, companies are preparing to spend billions of dollars on law firms’ internal records, business documents and videos of everyday physical tasks.
This is the focus of Business Insider’s new series, “The AI Data Grab,” which published three articles this week on the accelerating race to acquire workplace data. Taken together, the three reports show that AI development is moving beyond large models and faster chips toward understanding how work is actually performed, turning the ways humans work in the real world into a crucial raw material for the next generation of AI systems.
Law firms are turning decades of expertise into AI systems
Major law firms are building AI systems around their deal histories, litigation strategies and client information in the hope of gaining an advantage over general-purpose tools. Kirkland & Ellis has agreed to pay Palantir 500 million dollars to develop technology for its private-equity advisory business, while Latham & Watkins has bought its own Nvidia servers to customise large language models for its needs.
At Morgan & Morgan, America’s largest personal injury firm, lawyers use internal AI models trained on case patterns. The firm has committed to investing one billion dollars in the effort over the next decade. Chief Transformation Officer Yath Agarwal said: “We have more data than any other institution.” He added that the firm should “be the best steward of it.”
Harvey, LexisNexis and Thomson Reuters are also moving toward models built on firms’ own data, where LexisNexis Global Legal Chief Sean Fitzpatrick says law firms themselves requested the change.
AI labs are paying heavily for Slack messages and internal documents
The investigation found that AI labs are offering businesses six- and seven-figure sums for anonymised Slack messages, emails and internal documents, often accompanied by “exploding offers” with 48-hour deadlines. Dean Nicolls, CEO of data broker firm Mode.Inc, said: “Demand right now is extraordinarily high.”
According to Gartner, annual spending on AI training data is projected to rise from less than one billion dollars last year to nearly 5.6 billion dollars by 2027. The trend also has a darker side: when employees’ workplace data is sold, they rarely know about it or benefit from it.
Springbok CEO Doug Kruzkemp said he learned through the media, with less than 48 hours’ notice, that Google had bought his company’s data. Emory University law professor Ifeoma Ajunwa warned that, in the absence of federal privacy laws, the practice raises consent questions that remain unresolved.
Companies are paying workers to teach robots physical tasks
A third report examined how AI companies are paying people to record themselves performing everyday physical tasks in order to train robots. Shift, the data-collection arm of German startup Microagi, offered private chef Aliser Demir his usual rate of 100 to 150 dollars per person to cook while wearing a camera-equipped hat.
Demir said: “Since I started this job, I have been feeding people. Now I am feeding AI too.” According to Encord’s Vinet Welorugan, demand for robot-training data has increased roughly tenfold compared with a year ago, with clients now requesting up to 100,000 hours of annotated footage instead of 1,000 hours.
According to PitchBook, robotics companies raised 4.33 billion dollars globally in the first half of this year, surpassing the 7.28 billion dollars raised in the whole of 2025.