AI: Rush for Business Data, Growing Concerns

⚡

Key Takeaways

1AI labs are purchasing internal archives like Slack or Google Docs to train their models

2Six-figure offers with 48-hour deadlines target companies, while brokers anonymize and resell the data

3Annual spending on data for AI could reach $6.5 billion by 2027 according to Gartner

4Concerns are emerging about privacy, employee consent, and the lack of federal laws

💡Why it matters — The growing monetization of business data fuels AI training but raises major issues of security, consent, and regulation.

An emerging market is purchasing internal archives — from Slack messages to Google documents — to feed AI models. Amid expedited offers, rising budgets, and buyout programs, the collection is intensifying. However, it faces alerts regarding employee consent, data security, and the lack of federal safeguards.

Alarm Signals on Privacy and Security

Few employees benefit from the sale of their messages or know that they are being monetized, even as their Slack exchanges, presentations, and documents are among the coveted data sources. Illicit offers also exist: a stolen source code from a major company was reportedly offered for millions, according to a model training professional. An attempt at internal data collection at a large tech company sparked a backlash from employees. Legally, the absence of dedicated federal privacy laws leaves few barriers to the use of employee data for training AI. In this context, the expansion of these practices could accelerate calls for new protections, while reinforcing the idea that the value of daily digital traces could become a bargaining chip for workers.

Political Pressures Surrounding Spirit Airlines Data

The proposed purchase by a tech giant of old data from a bankrupt airline for $10 million was postponed partly due to concerns from flight attendants. Last week, Elizabeth Warren and over 100 members of Congress urged the two companies involved to protect the privacy of thousands of former employees of the airline. This operation illustrates the sensitivity surrounding the sale of archives from struggling companies.

Brokers and Acquisition Tactics: Tight Deadlines and Anonymization

A chain of intermediaries targets data-rich but cash-strapped organizations, promising high amounts in exchange for quick decisions. Offers with 48-hour deadlines are multiplying and can go as high as double the market rate, according to a broker who compares the situation to an oil boom. Before resale to model developers, datasets are cleaned of sensitive elements like social security numbers and addresses, then sold to buyers seeking an advantage.

⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Explosive Demand Driven by AI Agents and Major Players

Demand is described as astronomical by data sellers, as major groups are already exploiting information from their own employees while brokers purchase streams from active companies. The development of AI agents for businesses by OpenAI, Anthropic, Meta, and others increases the value of work data considered mundane. In this context, six-figure offers with deadlines under 48 hours are circulating among small business leaders, and advertising campaigns tout potential gains in the millions.

Concrete Case: Springshot Discovers Resale via a Third Party

A leader of a coordination tools provider learned in August, during an online call, that a digital giant had acquired data related to his business. He claims to have had less than 48 hours of media alert, while the intellectual property in question was built over 15 years. The company operates in airport professions, from gate agents to cleaning, illustrating the granularity of the sought-after datasets.

Prices, Desired Formats, and Announced Budgets

Some arrangements offer financial compensation ranging from $100,000 to $2 million for access to anonymized data from medium-sized companies. Training specialists place great importance on recordings that concretely illustrate how work is performed, deemed valuable for training AI agents, and prioritize the collection of longitudinal data series. Without access to sources like Slack, CRM, or emails, the value of a dataset decreases, according to a sales manager. Spending on data for AI was under $1 billion last year and is projected to approach $6.5 billion by 2027. Startup leaders see data licensing as a future widespread revenue source, estimating that 10% of the U.S. GDP could soon be dedicated to training models. Last week, a sector player announced a $1 billion commitment over 12 months to acquire business data. Typical datasets include Slack messages, presentations, and collaborative documents.