Fraud prevention for the agent economy

Humans.Agents.Attackers.

To your fraud stack, they look identical. xkeeper tells them apart — scoring every actor at your gate, and showing you exactly why.

live — the gateclick a row to inspect
why this score#—
actor
principal
credential
event
0 / 100
    policy — tuned by you

    Illustrative data — in production, this is your traffic.

    0%
    year-over-year growth in traffic from AI agents and agentic browsers
    HUMAN Security · 2026 State of AI Traffic Report
    0%
    of companies hit by attempted or successful AI-powered fraud in the past yearKPMG, 2026
    $0.00
    what every $1 lost to fraud actually costs a merchant, all-inLexisNexis True Cost of Fraud, 2026
    2 in 3
    US merchants worried about fraud from agent-driven commerceLexisNexis True Cost of Fraud, 2026

    The problem

    “Human or bot” is the wrong question.

    Legacy fraud tools sort traffic into two buckets — humans get through, bots get blocked. Here’s how that verdict goes wrong, three ways.

    Human

    The classic customer

    Still most of your traffic — still gaming promos and referrals, at human speed.

    legacy verdict✓ trusted human
    realityreferral ring · 3rd account
    Delegated agent

    Your customer’s agent

    Shopping and redeeming on a real customer’s behalf. Real intent, real revenue.

    legacy verdict✗ blocked as a bot
    realityyour customer · $184 sale
    Hostile automation

    An attacker at machine speed

    Malicious automation that looks human to legacy tools — card testing, promo farming, account takeover.

    legacy verdict✓ passes as human
    realitycard testing · 63 cards

    And the patterns won’t hold still.

    Block a pattern today and it mutates by next week. Fraud has always been a cat-and-mouse game — and the mouse now has AI.

    So xkeeper puts a keeper agent on watch — surfacing new patterns as they emerge, then reporting, acting, or adapting your rules as the game shifts. You set how much it does on its own.

    The questions that matter now: who is this actor, who do they act for, what are they allowed to do — and what will they try next? That takes a fraud engine built to ask, answer, and adapt.

    The product

    Built as a glass box.

    Fraud tools ask you to trust a number. xkeeper shows its work — every score explained, every factor tunable, every agent action logged.

    /01

    Every score ships its reasons

    Each decision carries a plain-language factor breakdown — what contributed, by how much, and why. Audit-ready by default, for your team and your regulator.

    /02

    Factors you can tune

    Your risk tolerance isn’t ours to decide. Reweight the factors that matter to your business, version your policies like code, and roll back when an experiment doesn’t pan out.

    /03

    A model per problem

    Payment fraud, promo abuse, account takeover, one specific campaign — run separate models per use case or cohort instead of forcing one score to mean everything.

    /04

    Rules on any event

    Bring your own events. Trigger rules on anything you can send us — a signup, a redemption, a custom signal only your product emits.

    /05

    Every actor leaves fingerprints

    Devices, cards, emails, phone numbers — shared fingerprints connect “unrelated” accounts into rings. xkeeper maps every connection, and keeper agents walk the graph at machine scale, finding clusters no human could piece together case by case.

    /06

    Keeper agents on watch

    Task an agent to monitor a cohort, surface emerging patterns, file reports — or adapt your rules and act, with autonomy you dial up or down. Every action logged, every conclusion explained. 24/7.

    keeper agent · pattern watchpromo cohort · illustrative

    Developers

    Live before lunch.

    One SDK, one call per event. Send us the request — with whatever signals ride along: agent credentials (KYA), checkout-protocol context (ACP), device, your custom fields. Back comes a score, an action, and the reasons — as data, not a dashboard screenshot.

    No sales call to see the product. No six-month integration. Docs, keys, and a sandbox from day one.

    $ npm install @xkeeper/sdk
    
    import { xkeeper } from "@xkeeper/sdk";
    
    const decision = await xkeeper.evaluate({
      event: "promo.redeem",
      actor: req,          // headers, device, credential
      context: { code: "SPRING24" },
    });
    
    decision.action   // "block"
    decision.score    // 94
    decision.because  // ["credential_fails_verification",
                      //  "redemption_velocity_22x_cohort",
                      //  "device_linked_to_41_accounts"]

    Use cases

    One engine. Your fraud, specifically.

    Payment fraud

    Card testing, stolen credentials, chargebacks — at human and agent speed.

    Promo abuse

    Referral farming, coupon stacking, signup bonuses milked by fleets of agents.

    Account takeover

    Credential stuffing and session hijacking, including agents with stolen delegation.

    Campaign abuse

    Drop-day bots, bonus hunters, one-campaign scams — spin up a model for the moment, retire it after.

    Content fraud

    Fake reviews, spam listings, and synthetic engagement at generative scale.

    Early access

    Get in before the agents do.

    The agent-payment rails made one thing explicit: the fraud decision is still yours. We’re onboarding a small set of design partners — tell us what’s hitting you hardest and we’ll be in touch.

    You’re on the list.We’ll reach out as design-partner slots open. Thanks for telling us what hurts — it decides what we build first.