Updated 2026-06-30 · category: payment-risk / agent-guardrails

x402 payment-risk oracles, compared

Q: What is a payment-risk oracle?

A payment-risk oracle is a service an AI agent calls before authorizing a payment, to get an independent risk verdict on the counterparty and the amount. Blackwall is a payment-risk oracle for the x402 protocol: it runs in the pre-signature window and returns GO / HOLD / STOP before the agent signs.

Q: How does Blackwall detect a price gouge in an x402 payment?

Blackwall compares the quoted amount to the counterparty's own median price for that resource, computed per distinct payer so it is wash-trade-hardened. A STOP verdict fires when the amount is at least 8x that median.

Q: Does Blackwall take custody of funds?

No. Blackwall returns a verdict and is never in the settlement path. Your agent decides what to do; funds are never held or moved by Blackwall.

If your AI agent pays for things over x402, something should decide whether a given payment is safe before it signs. Several services now answer a pre-payment question — but they judge different things. Here is what each one actually checks.

What is a payment-risk oracle? A service an AI agent calls before authorizing a payment, to get an independent risk verdict on the counterparty and the amount. For x402, a payment-risk oracle runs in the pre-signature window — after the 402 Payment Required challenge, before the agent signs — and returns a GO / HOLD / STOP style verdict. Full definition →

At a glance

	Blackwall	Ontario Protocol	KYT / compliance (AnChain, Chainalysis…)
Question it answers	Should my agent trust this payee with this payment, at this price?	Is this endpoint set up correctly and within my budget?	Is this wallet sanctioned or tied to illicit flows?
Verdict	GO / HOLD / STOP + signed receipt	allow / review / deny (free `can-pay`)	screen result / wallet risk score
Runs pre-signature	Yes	Yes	At compliance screening
Counterparty payment behavior	Yes — Bayesian, from on-chain settlement history (dispute signal pending outcomes)	Reputation via EAS attestations (declarative)	No
Price fairness / anomaly	Yes — vs the counterparty's own per-payer median; STOP at ≥8×	Price-aware vs your budget, not a fairness baseline	No
Sanctions screening	Yes — OFAC sanctioned-address list, auto-refreshed (best-effort)	No	Yes — this is its core
Endpoint readiness	Yes — folds a self-scored grade in	Yes — this is its core	No
Takes custody of funds	No — verdict only, never in the settlement path	Advisory; optional settlement proxy	No
Interface	HTTP API + MCP (`forecast_payment`)	HTTP API + MCP	API (some via MCP)
Cost	x402 pay-per-forecast; free under $10 at risk	Free `can-pay`; paid listings + scans	Enterprise-priced

These are layers, not rivals

The useful way to read the table is as a stack, not a bake-off. An agent about to pay an unknown endpoint has three separate questions, and each service owns one:

Is the endpoint even set up right, and within budget? — endpoint-readiness (Ontario Protocol).
Is this payee trustworthy with this payment, at this price? — counterparty payment-risk (Blackwall).
Is the wallet sanctioned? — compliance / KYT (AnChain-class).

They compose. Blackwall can even take a readiness grade as one input to its verdict — so "use both" is the honest recommendation, not "switch." Where Blackwall is different from the nearest neighbor is signal depth: it judges the counterparty's real settlement and dispute behavior and whether the price is fair, which the readiness-and-budget check does not.

See a verdict, live

Blackwall is live on Base (mainnet-configured). You can get a verdict without a payment for amounts under the free threshold:

curl -sS -X POST https://agent-egress-proxy.onrender.com/v1/forecast-payment \
  -H 'content-type: application/json' \
  -d '{"counterparty":"0x…","amount":"5.00","asset":"USDC","chain":"base"}'
# → {"verdict":"STOP", "reasons":["quoted amount is 10.0x the counterparty's own median …"], …}

Discovery descriptor: /.well-known/x402. MCP tool: forecast_payment (stdio).

FAQ

What is a payment-risk oracle?

A service an AI agent calls before authorizing a payment, to get an independent risk verdict on the counterparty and the amount. Blackwall is a payment-risk oracle for x402: it runs in the pre-signature window and returns GO / HOLD / STOP before the agent signs.

How do I check the risk of an x402 payment before signing it?

Call a pre-signature payment-risk oracle after you receive the 402 and before you sign. With Blackwall, POST the counterparty, amount, asset and chain to /v1/forecast-payment and act on the verdict: sign on GO, escalate to a spending cap or human on HOLD, do not sign on STOP.

Blackwall vs Ontario Protocol — what's the difference?

Both return a pre-payment verdict at the same moment. Ontario judges endpoint readiness and your budget cap; Blackwall judges the counterparty's settlement and dispute behavior and whether the price is fair. Different evidence, and complementary — Blackwall can take Ontario's readiness grade as one input.

How does Blackwall detect a price gouge?

It compares the quoted amount to the counterparty's own median price for that resource, computed per distinct payer so it is wash-trade-hardened. A STOP fires when the amount is at least 8× that median.

Does Blackwall take custody of funds?

No. It returns a verdict and is never in the settlement path. Your agent decides; funds are never held or moved by Blackwall.