Released 2026-05-30. Content moderation now covers public forms, and the four moderated surfaces share one engine with per-surface evaluator control.
Form content moderation
- Public forms are screened against the platform policy — title, description, success / response message, section titles, field labels, dropdown options, consent text, and welcome / success screen copy — on create, on save, and by a background sweep. See Form content moderation.
- A high verdict automatically takes the form off public surfaces: it is set private and embed is disabled, and the owner is emailed. The publish gate refuses to make a blocked form public until the flagged content is fixed (then the AI re-checks and can clear it) or an admin lifts the block.
- Admins get a Form moderation queue with block / approve / re-evaluate actions, mirroring the workflow / agent / playground queues.
- Form submissions (visitor input) are not screened by this check — only the owner-authored form content.
One moderation engine, per-surface evaluator
- The generic content classifier (taxonomy, AI prompt, blocklist scanner) is now a single shared module used by workflows, agents, playgrounds, and forms — so the rules and categories stay identical everywhere.
- Admins can choose the evaluator per surface independently: the rich AI classifier (uses platform tokens) or the fast deterministic blocklist (free, no network). The OpenAI model, blocklist, fail-open, and sweep cadence remain shared policy.
- The admin Moderation settings page (formerly "Workflow moderation") now exposes a per-surface evaluator picker and a form-moderation toggle.
Clearer moderation emails
- Block, review-request, and decision emails now name the actual surface ("Your form has been blocked…", "HIGH-risk agent auto-blocked…") instead of reusing a workflow-shaped template with a prefixed name.