AI Is Being Trained on Your Secrets – What You Need to Know

Founders of bankrupt startups are selling former employees' work correspondence to AI developers. Corporate data from Slack and email is becoming training material for models. Experts warn of privacy risks and the quality of anonymization.

Amid the AI boom, a new market has emerged: founders of bankrupt startups are selling corporate archives of their former employees to artificial intelligence developers. Corporate correspondence, emails, and Slack messages are becoming valuable material for training new AI models that need realistic examples of work processes. This is reported by Futurism. Based on the correspondence, RL trainers are created—virtual work environments where AI learns to perform office tasks. Interest in the data is high: Anthropic plans to allocate about one billion dollars to develop training platforms this year. After closing their businesses, entrepreneurs sell archives through specialized intermediary services. One platform introduced a tool for data anonymization; over the past year, it facilitated nearly a hundred deals, and startup owners received over a million dollars. Experts warn of privacy risks: if office conversations were recorded and sold, it would cause a public outcry. Corporate messengers often contain personal information. Mark Rotenberg, founder of the Center for AI and Digital Policy, stated there are serious confidentiality issues. The quality of anonymization is also questionable: if data is not sufficiently hidden, developers will be able to identify employees, increasing the risk of leaks.

AI Is Being Trained on Your Secrets – What You Need to Know