Microsoft announces Copilot+ PCs and Recall
Recall continuously screenshots the desktop to build a searchable history; security researchers found the local database unencrypted within days, and Microsoft delayed the rollout.
- Culture & impact
- Security & misuse
- Notable
Microsoft announced Copilot+ PCs, a new category of Windows laptops built around a dedicated neural processing unit capable of at least 40 trillion operations per second, and unveiled Recall as its flagship feature: a tool that periodically screenshots everything on screen and lets users search their own activity history in natural language — “photographic memory” for the PC, in Microsoft’s framing.
Recall worked by storing snapshots locally and indexing them by content and context, so a user could scroll back through time or search for something they remembered seeing rather than the file, app or website it came from. Microsoft said snapshots stayed on-device, that users could delete individual entries, exclude apps or time ranges, and pause capture entirely. The launch bundled other generative features — Cocreator for AI image generation in Paint and Photos, live translated captions across more than 40 languages, and an integrated Copilot accessible from a dedicated keyboard key — with pricing starting at $999.
Recall drew immediate security criticism once outside researchers examined how it worked. Within days, security researchers demonstrated that the local database of screenshots was stored without encryption and could be extracted by any process with local access, including malware, effectively creating a searchable log of everything a user had ever seen on screen — passwords, private messages and financial details included — in one accessible file. Microsoft had planned to ship Recall as an on-by-default preview feature on 18 June; instead it delayed the release, made the feature opt-in rather than default, and added encryption and authentication requirements before eventually shipping a reworked version months later.
The episode became a reference case for the gap between a capability demo and a capability shipped to hundreds of millions of machines without adversarial review, at a moment when several major platforms were racing to embed AI features directly into operating systems.