Your AI writes the site. We keep it online.

Archives are unpacked before publish; path traversal and decompression bombs are rejected. The page URL and every link in its HTML and SVG are checked against Google Safe Browsing, and a hit refuses the publish with a 422. Raster images and sampled video frames are classified for explicit content by OpenAI. That classifier is advisory: it flags for review, it never blocks a publish.

Cloudflare's CSAM Scanning Tool only hashes images entering the edge cache, which a dynamic Worker response would bypass, so we deliberately write servable image bytes into the Cache API. A confirmed detection suspends that end-user at once and fires platform.subject.csam_suspended. The page is withheld with 451.

Three signals withhold a page, in fixed order: our moderation status, an admin lock on one of your end-users, then a plan lock for non-payment. Withheld pages return 404. Payment recovery clears only the plan lock, so it can never restore content withheld for abuse.

One PATCH, or one click in your dashboard, stops a customer publishing or editing. Sites already live keep serving: a write freeze is not a takedown. To pull content, delete the site by id. Your Customers view carries the roster, stats, a file inspector, edit history and abuse reports. Staff it with contractors if you like: the support role sees customers and nothing else, and cannot reverse one of our abuse actions.

We score each platform on the share of active end-users with a confirmed adverse action over 90 days: under 0.5 percent normal, 0.5 advisory, 2 probation, 5 suspension. Three confirmed CSAM detections reach probation on their own. Under 20 active subjects, rate alone never passes advisory. Every change fires platform.abuse.tier_changed.

For safety and abuse processing we are an independent controller, not your processor. That is what lets us act on illegal content without waiting for an instruction from you.