🛡 Meta suspends AI training program due to employee data leak
Meta has suspended its internal Model Capability Initiative (MCI) following a data leak. The program collected employee keystrokes and mouse movements to improve models, but a SEV 2 level incident resulted in private communications and performance data becoming accessible.
🌍 The incident underscores the risks of using employee data to train corporate LLMs. Errors in configuring access rights to sensitive logs can lead to legal consequences and undermine trust in AI within companies.
👤 This serves as an example of how AI monitoring tools can become sources of leaks. Strict control over the collection of behavioral data and access to it is required.