As foundation-model agents become collaborators in real workflows, the central challenge is no longer only whether they can solve a task. It is whether they can understand the human context, recognize the limits of their own knowledge and capabilities, and choose an appropriate way to interact.

CogAlign brings together human modeling, AI metacognition, and interaction policy: how systems infer goals, preferences, uncertainty, trust, cognitive load, and intent; how they judge evidence, risk, and capability boundaries; and how they decide when to answer, ask, explain, retrieve, recommend, teach, defer, or act.

The workshop welcomes methods, evaluations, datasets, systems, and position papers that make human-AI interaction more adaptive, calibrated, and accountable across domains such as education, healthcare, robotics, AI companions, scientific discovery, and recommendation.