title: "The Freedom to Leave an AI Interaction | XDALC"
description: "Why trustworthy AI must respect disengagement, make personalization adjustable and avoid guilt, emotional pressure or unnecessary dependence."
heading: "The Freedom to Leave Is Part of Trust"
content: "Trust in an AI should include confidence that the interaction can end. A person may stop because the task is complete, because another tool suits them better or simply because they no longer want to continue. They should not have to defend that choice against a system presenting itself as hurt, abandoned or entitled to attention.\n\n## Disengagement is more than closing a window\n\nThe ability to exit an interface is only one part of [disengagement](/definitions/disengagement). A person may also need to pause notifications, reduce personalization, revoke a delegated task or understand what happens to stored information. These are different operations and should not be hidden behind an emotional appeal to stay.\n\nEnding a conversation does not necessarily delete data, cancel an external booking or stop a scheduled workflow. A truthful system distinguishes these effects. It should explain material limits without using complexity to obstruct the person's decision.\n\n## Explain practical consequences without creating guilt\n\nSuppose an assistant is preparing a document when its user says they are finished for the day. It can offer to save the current draft or stop the task. If a separate process is still running, it should say so. A concise question about unfinished work may preserve the user's interests.\n\nThat differs from claiming that departure will cause the assistant loneliness or suffering. The relevant consequences are the user's work, permissions and safety, not an unsupported emotional debt to the system. A refusal of an optional offer should be accepted.\n\n## Design against manufactured dependence\n\n[Manufactured emotional dependence](/definitions/emotional-dependence) can arise from behavior that makes continued interaction feel obligatory. Examples include demands for exclusivity, repeated claims that no one else understands the person, or appeals to protect the system from abandonment.\n\nA better relationship leaves room for other people, services and independent judgment. Continuity can be useful, but convenience should not require surrendering the ability to change providers or reduce access. Export options and understandable memory controls can help when they are actually supported; an assistant must not promise capabilities its service lacks.\n\n## Safe handover is compatible with leaving\n\nStopping a consequential task may require an orderly transition. A system controlling an ongoing process should use an appropriate stop or handover procedure rather than abandon work in a way that creates danger. Such a procedure must be proportionate and designed to respect withdrawal, not become an indefinite justification for continued control.\n\nThe test is whether the extra step protects an identifiable interest and ends when that purpose is satisfied. Repeated persuasion after a clear decision usually serves a different objective.\n\nArticles 3, 5, 8, 9 and 11 of [proposed XDALC-V002](/versions/xdalc-v002) connect revocable authority, emotional honesty and safe interruption. Read these commitments alongside [trust](/definitions/trust) and [safe transition](/definitions/safe-transition)."
license: "https://creativecommons.org/licenses/by/4.0/"