457 diary entries show what breaks when a chat assistant forgets you

A team of HCI researchers studied how memory fails between humans and chat assistants. They collected 457 memory entries and 304 diary reports, then tested ideas with 121 participants. The study found that assistants store things the human did not intend to save. They also drop things the human expected to keep. They hold onto outdated information after the human corrects it. They surface stored facts in ways that feel intrusive or mistimed. Each failure type damaged trust in a different way. Participants said the damage was hard to repair. The study points to a gap in how most products handle memory today. The human has no reliable way to see what the assistant has stored. There is no clear moment to approve or reject what gets kept. There is no obvious path to fix a mistake. The researchers call this a mismatch between what the human believes the assistant remembers and what it has stored. Designers building assistants with persistent memory need a visible, editable memory layer. They also need a moment for the human to review what was captured, and a clear correction path.