LangGraph 1.2.20 fixes interrupt response mismatch in human-in-the-loop

LangGraph 1.2.20 shipped on 5 November 2026 with two changes that look, on paper, like plumbing: the framework now assigns every interrupt() in a task its own id, and it adds a strict resume mode that refuses stale interrupt answers.

On 10 November 2026, a writer on DEV Community's AI section argued that the two lines are actually a correctness fix addressing one of the nastiest failure shapes in human-in-the-loop workflows. In the failing scenario, a person reviews an agent's proposal and clicks Approve, but the response gets applied to a different question, one that has already shifted since the interrupt was raised. Separate ids per interrupt make a resumed answer target a specific, identified interrupt rather than whatever currently matches a pattern; strict resume then discards answers whose underlying interrupt no longer matches, rather than consuming them silently.

Verification is limited to this single piece, so no earlier presentation or thread is documented here and no follow-up behavior is described. The material fixes the record at the release itself and the community reading of it: official changes in 1.2.20, one observer framing them as a fix for misrouted approvals.