Overview
- Reporting published on Monday and Tuesday presented internal screenshots and documents that showed full prompts, transcripts, private staff chats and personnel and performance records were broadly accessible inside Meta.
- The tool at the center of the reports is the Model Capability Initiative, introduced in April to capture employees' keystrokes, mouse movements, clicks and related telemetry to train internal AI models.
- Meta confirmed it has suspended the program while probing the issue, said it has found no evidence so far of employees misusing the data, and described the pause as being rolled out in stages.
- Internal materials shown to reporters flagged the incident as 'SEV 2' on an internal 1–5 severity scale, and screenshots captured angry employee messages about promised protections not being enforced.
- The episode follows other recent Meta security problems and is likely to raise fresh questions about employee privacy, the company's operational controls and potential regulatory or operational consequences.