What happened
Researchers developed a method for removing specific data from trained AI models while preserving the accuracy of similar data the model should keep — a problem that was harder than it seemed because related information gets tangled together during training. This matters because it could let companies actually comply with data deletion requests (like the EU's right to be forgotten) without retraining their entire system from scratch.
Why it matters
If this works reliably, it removes a major practical bottleneck in privacy regulation: currently, deleting a person's data from an AI model either requires retraining from scratch (expensive, slow) or doesn't actually work (the model still remembers). A fast, reliable unlearning method would make privacy deletion legally and economically feasible at scale.