Meta Logged Every Keystroke for AI Training, Then Left That Pile Open to Whole Company
Meta's Model Capability Initiative logged keystrokes, clicks, and screenshots from staff laptops to train AI, and 1,600 employees who warned it was a security risk were proven right when the data sat open to everyone inside the company.
"no completed privacy reviews were provided"
Wired, via Pivot to AI (David Gerard)
Meta launched the Model Capability Initiative in April, and the tool quietly collected keystrokes, mouse movements, and screen content from US employee laptops to train AI systems to use software the way humans do. More than 1,600 Meta employees signed an internal petition demanding it be halted, citing privacy and data security concerns, and the company answered with vague safeguards and an opt-out it reserved for executives. Then the punchline arrived on its own: Meta left potentially sensitive information collected from employee laptops accessible to anyone inside the company, paused the data collection, but wants to start keylogging again as soon as possible. A surveillance machine that leaks the surveillance is the most honest product Menlo Park has shipped in years. Pivot to AI on Meta’s keylogging leak
Source: Wired, via Pivot to AI (David Gerard) · David Gerard