Version history.

Versions

• 2026-10-08Section 02 adds seven rules: no using the service or its output to build a competing product or train a competing model; no bulk harvesting of output to copy or distil a model; no deceiving voters or impersonating candidates and officials; no content that encourages or glorifies suicide, self-harm or eating disorders; no sexually explicit content; no attacks on critical infrastructure; and no passing AI output off as human-made to deceive someone. It also says plainly what the automatic text filter checks and what it does not, and that shared pages are not labelled as AI-generated. Current version
• 2026-09-27Appeals can be filed in the product or through the appeal form, each with a reference, as well as by email. This version applied until the version dated 2026-10-08 replaced it on this site. Read this version
• 2026-09-22The website now starts on Skip approvals instead of asking before every tool action, and the policy describes the actions that still need your review. This version was settled on 2026-09-22 and replaced on 2026-09-27 before it was published on this site; the first version published here after it is dated 2026-09-27. Read this version
• 2026-08-05The earliest recorded version. This version applied until the version dated 2026-09-27 replaced it on this site. Read this version

A version’s date is the day its text was settled, which can be earlier than the day this site published it. Earlier versions are shown for reference only; the current version is the one that applies.