Meta Halts Internal AI Training Programme After Employee Data Leak

Menlo Park: Meta has suspended an internal AI training programme that tracked employees' computer activity, after private conversations and activity data belonging to staff became accessible to others across the company.

Although the programme was claimed to comply with security protocols, the data leak has triggered a major controversy. The initiative shut down was called the "Model Capability Initiative," developed by Meta's Superintelligence Labs team.

The programme's goal was to study how humans interact with computers in order to improve the capabilities of AI models. As part of this effort, data including employees' keystrokes, mouse movements and clicks were being collected.

This data — including employees' private conversations and performance-related information — ended up being openly visible to others within the company. Business Insider first reported the leak.

Employees expressed sharp anger over the breach of the security assurances given when the programme was launched. The protest stemmed from the fact that the lapse occurred despite assurances that the data would be kept secure.

A Meta spokesperson said there is currently no indication that the data was misused by anyone. However, the programme has been temporarily suspended pending an investigation into the incident. The initiative, originally launched under the name "AI for Work," had later been rebranded as the "Agent Transformation Accelerator."

The programme, which was mandatory for most employees, had faced opposition from staff from the very outset. With privacy concerns already simmering, the data leak has now become a significant setback for Meta.