Evidence record / paperclip

A dated record of one change, kept so the writing that cites it can be checked. Compiled from the sources listed below by the research run, not written for reading. The judgment lives in the issues that cite it, below.

2026-10-13-paperclip-v2026-1005-0-lets-an-agent-record-a-user-chat-reply-as-a-confirmation-decision

v2026.1005.0 lets an agent record a user chat reply as a confirmation decision. An active responding run can resolve a confirmation by citing the latest user message from the conversation user after the card. The server records the agent interpretation and evaluates the resolver policy as that agent; the comment is provenance, not server-verified consent. Governed actions must retain their dedicated approval mechanisms. Under anyone the creator can resolve its own card. human_only requires a human resolver; not_creator bars only the creator and can still allow another eligible agent. v2026.1001.0 also queued answers and approvals during active runs rather than bouncing them (#13539).

Channel: tagged-release. Half: both. Date: 2026-10-06.

Operator consequence: Use human_only where a human resolver is required, and retain dedicated spend/tool/hiring approvals for governed actions. Use not_creator only where creator self-resolution is the boundary you intend. In ordinary chat, this lets a person answer once instead of repeating a click; audit the agent decision through the recorded comment ID. The trade is consent interpretation by the model, not conversion of that comment into a human credential.

Receipt

Finding metadata

Run: 2026-10-13-profile-refresh-2026-09-21_2026-10-13-frontier-v0

Finding ID: 2026-10-13-paperclip-v2026-1005-0-lets-an-agent-record-a-user-chat-reply-as-a-confirmation-decision

Profile citations

Source links

Primary links, including exact changelog lines when available.

Versioned source: run artifact