Meta has paused an internal effort to train AI models using employee keystroke data, according to the company’s CTO. The pause follows backlash over the “Model Capability Initiative,” which began only two months before it was halted, the outlets report. The CTO says the data was “put in a place it wasn’t supposed to go,” indicating the information may have been stored or handled in an unintended system or environment.
The reporting describes the pause as temporary and implemented shortly after the initiative triggered scrutiny. Meta has faced criticism from some employees and external observers over how workplace data is collected, stored, and used in AI development. While the company’s broader goals for building and improving AI models remain unchanged, the pause reflects an internal reassessment of the specific data source and the processes around it.
Across the accounts, the main points are that Meta is stopping the keystroke-data training component, that the decision is linked to a concern about improper handling or placement of the data, and that the initiative had been running for a brief period before being suspended.
