First test to measure whether AI systems actually remember context across conversations

Researchers built the first standardized test to measure whether AI systems can reliably remember and retrieve facts from past conversations without mixing them up. This matters because AI companies have added memory tools (vector databases, long context windows) but nobody could actually verify they work — now someone can.

Right now, companies claim their AI systems have 'continuity' or 'memory,' but there's no agreed way to measure whether that's true or marketing. This framework gives you a ruler. The test runs 250 different life stories through the system and checks whether it retrieves the correct fact for the correct person without cross-contamination — a deceptively hard problem that degrades as the database grows. If this gets adopted, it stops being possible to ship memory systems without proving they actually work at scale.