The Meta report adds another layer to an already tense period inside the company, where AI development, restructuring, and workplace surveillance concerns are all colliding.
At the centre of it is an internal program called the Model Capability Initiative (MCI), where employees’ on-device activity — including mouse movements, clicks, keystrokes, and occasional screen snapshots — was being used to train internal AI systems. According to the report, a configuration issue or internal exposure meant that sensitive employee data (including performance-related information and chat-like internal content) became more widely accessible within Meta Platforms than intended.
Meta classified the incident as SEV 2 (a serious but not top-level crisis) and has temporarily paused the program while investigating. The company has also publicly said there’s no current evidence that the data was improperly accessed, but the exposure alone was enough to trigger an internal shutdown and review.
What makes this particularly sensitive is the context. The program itself was already controversial because participation was reportedly mandatory for many employees, and workers were uneasy about being continuously monitored as part of AI training. The leak essentially validates those concerns, even if no malicious access has been confirmed.
This also sits on top of broader internal strain. Leadership has already acknowledged weak morale amid layoffs, shifting roles, and aggressive restructuring around AI priorities. So the incident isn’t just a technical issue — it feeds into a larger trust problem between employees and the company’s AI strategy.
Stepping back, this reflects a wider trend across tech firms: as companies race to build more capable AI systems, they are increasingly collecting granular behavioural data from employees and users. That creates a tension between “training fuel” for AI and workplace privacy expectations, especially when data collection becomes continuous or opaque.
If Meta’s investigation finds that safeguards were insufficient rather than just misconfigured, this could push the company — and others watching closely — to redesign how internal AI training data is collected, or at least tighten separation between operational employee data and model training pipelines.
