Skip to content

Use case · AI credit farming

Stop bleeding GPU dollars to credit farms.

Every AI generation tool fights the same math: $0.10/VM rents an account that resells $5–$30 of inference. Root Herald raises the cost of an account to the cost of a real chip, and the arbitrage collapses.

Economics

Yield meets the hardware floor.

Per-fake-account yield is the resale value of extractable credits: $5–$30 on typical platforms. The post-Root-Herald floor is $30+ per real chip. The crossover happens at the low end of the yield distribution and dominates above it.

AI free-tier credit farming

Suno, Lovable, Krea, Civitai

Solved
$30 floor

Per-identity yield

$5 – $30

Rational ceiling

< $10

Only Tier 1 was profitable; Tier 3+ exceeds yield.

The cases we close

What stops working under strict policy.

Cloud-VM fleets

NitroTPM, Azure vTPM, GCP Shielded: all rejected at chain validation. The default policy treats these as virtual and refuses.

swtpm in Docker

Trivial to spin up, trivial to reject: a software TPM can't chain its EK cert to a pinned manufacturer root, and its issuer DN labels it as an emulator.

Anti-detect browsers + proxies

They defeat fingerprinting, but the attestation is still grounded in the underlying hardware: one chip, one EKpub, one deviceId across every fake fingerprint.

Email / phone churn

Throwaway emails and SIM-farm phones don't generate fresh hardware. The deviceId stays the same across rotations.

Integration

One call at signup, one at credit-grant.

Attest at signup to gate account creation, then re-attest at each free-credit grant so takeover or device-share scenarios surface. The deviceId is stable per-tenant; bind it to the user record.

signup handler · @rootherald/nodets
import { RootHerald } from "@rootherald/node";

const rh = new RootHerald({ secretKey: process.env.RH_SECRET_KEY }); // rh_sk_…

// Earlier: rh.issueChallenge() -> relay nonce to your dumb client, which
// collects the opaque evidence and posts back { challengeId, evidence }.
const verdict = await rh.verify(evidence, {
  challengeId,
  policy: "rootherald:builtin:strict-hardware",
});
if (verdict.device.verdict !== "pass")
  return res.status(403).json({ error: 'device_check_failed' });

const existing = await db.users.findByDevice(verdict.device.ueid);
if (existing) return res.status(409).json({ error: 'device_already_registered' });
await db.users.create({ deviceId: verdict.device.ueid, email });

Real-world precedents

The industry signal.

Major AI providers have pulled back automatic free credits and banned third-party harnesses, citing credit-farming arbitrage that makes free promos unviable. Arkose Labssells an "LLM Platform Abuse" SKU; the category exists because the problem is widespread. Civitai, Lovable, Suno, Krea, and Janitor AI all sit in the named ICP for credit-farming pain.

Keep the free tier. Lose the abuse.

Free up to 10K attestations a month. One call at signup. The math just works.