How is this different from a content moderation API?

A moderation API reads the text and returns a policy category. That catches profanity and misses the scam, because a sophisticated scam contains no banned words. Rupt reads for intent, then weighs it against the device, the email and the network behind the account, which no text model can see.

Can it catch phishing pages our users deploy on our product?

Yes, and it is why app builders and hosting platforms run it. Rupt reads what your users ship, not just what they type, so a cloned bank login or a wallet drainer behind a launch page is caught before it goes anywhere. The operator's other accounts come down with it.

What surfaces can it screen?

Anything your users create. Listings, job and task posts, messages and chat, comments, reviews, profile bios, and the sites and apps built on your product. You set the bar per surface.

Does it work when a real human writes everything by hand?

That is the case it is built for. We have tracked operators who typed every message themselves, one new account a day for months. No bot check fires. Reuse is what gives them away: one browser carrying nineteen identities, one IP carrying eighteen signups.

Will it flag my legitimate users?

Rupt starts conservative and hands you the evidence, so a borderline item reaches your queue with its reasons rather than a takedown. Weighing content alongside device and network signals means an unusual message from a real user does not look like a scam.

Do we still need human moderators?

Most teams keep people in the loop and use Rupt to decide what reaches them. Obvious scams are held, clean traffic never enters the queue, and what is left arrives with the linked accounts and device history already attached.

Does it slow down my product?

No. Verdicts return in under a second, so you can hold an item on the verdict or publish and act straight after.

How long does it take to integrate?

Most teams ship in a day. Drop in the SDK, send the content you want read, and act on the verdict. If you already run Rupt, this is a policy change.

AI content moderation

1Detect

An AI that reads for intent, not for keywords.

A sophisticated scam contains no banned words. It is a plausible offer with an implausible way to pay, or a login page that is perfect apart from the domain. Rupt reads what the content is actually trying to get someone to do.

Moderation /Detections

Live

Caught in the last hour

ListingBelow market, wire transfer onlyScam

SiteBank login cloned on a subdomainPhishing

MessageDeposit first, then we move to WhatsAppScam

TaskUpload your passport to applyFraud

ReviewSame praise, nine listingsSpam

ListingTwo bed, available FridayClean

The operator, not the post

SignalThis oneNormal

Same browser fingerprint191

Same origin IP181

Reused throwaway number61

Items posted per day50+3

Naming pattern match700

Automation detectedNoneNone

What your policy did

HoldNever reaches your users

ReviewQueued with the evidence

ChallengeAuthor verifies by SMS

Link19 sibling accounts flagged

AllowClean author, no delay

surface covered: Every surface covered
to a verdict: <1s to a verdict
languages: 30+ languages

2Link

Tackle the root of the problem. Not a whack-a-mole approach.

Take one listing down and it is back tomorrow under a new account. So Rupt scores the source device, email, phone and network, and links them to every other identity the same person is running.

Moderation /Operator

Live

Caught in the last hour

ListingBelow market, wire transfer onlyScam

SiteBank login cloned on a subdomainPhishing

MessageDeposit first, then we move to WhatsAppScam

TaskUpload your passport to applyFraud

ReviewSame praise, nine listingsSpam

ListingTwo bed, available FridayClean

The operator, not the post

SignalThis oneNormal

Same browser fingerprint191

Same origin IP181

Reused throwaway number61

Items posted per day50+3

Naming pattern match700

Automation detectedNoneNone

What your policy did

HoldNever reaches your users

ReviewQueued with the evidence

ChallengeAuthor verifies by SMS

Link19 sibling accounts flagged

AllowClean author, no delay

identities linked: 19 identities linked
actor signals: 40+ actor signals
bot signals needed: 0 bot signals needed

3Act

You decide what a verdict is worth.

Hold content before anyone sees it, queue it with the evidence attached, make the author verify by SMS, or take down the whole ring at once. Set the bar per surface, so a first review is treated differently from a first payment request.

Moderation /Policy

Live

Caught in the last hour

ListingBelow market, wire transfer onlyScam

SiteBank login cloned on a subdomainPhishing

MessageDeposit first, then we move to WhatsAppScam

TaskUpload your passport to applyFraud

ReviewSame praise, nine listingsSpam

ListingTwo bed, available FridayClean

The operator, not the post

SignalThis oneNormal

Same browser fingerprint191

Same origin IP181

Reused throwaway number61

Items posted per day50+3

Naming pattern match700

Automation detectedNoneNone

What your policy did

HoldNever reaches your users

ReviewQueued with the evidence

ChallengeAuthor verifies by SMS

Link19 sibling accounts flagged

AllowClean author, no delay

actions available: 5 actions available
surface thresholds: Per surface thresholds
releases from you: 0 releases from you

Detect

Act

Free tools

By use case

By industry

Developers

Learn

AI content moderation.

Stop scams, spam and phishing across listings, messages, posts or entire sites.

What this looks like in production:

Built for platforms where users create for other users.

Marketplaces and job boards

Coding and hosting platforms

User to user communication

End-to-end content moderation, built for the AI era.

An AI that reads for intent, not for keywords.

Tackle the root of the problem. Not a whack-a-mole approach.

You decide what a verdict is worth.

State-of-the-art content moderation, and everything it normally ships without.

Content intelligence

Device intelligence

Email and IP intelligence

Progressive verification

Questions teams ask before turning on AI content moderation:

Find out what is already running on your platform.