# Model Attestation & Substitution Detection

> Model Attestation & Substitution Detection is a paid API for AI agents from dev-5090-box.tailfb4c1f.ts.net, paid per call via x402, $0.1/call, status unknown (last checked 2026-10-12).

Fingerprints or verifies an OpenAI-compatible inference endpoint to detect silent model substitution by the seller

## Facts

- Endpoint: POST https://dev-5090-box.tailfb4c1f.ts.net/v1/attest/model?utm_source=zero.xyz
- Price: $0.1/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-12
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/model-attestation-substitution-detection-5646d4f8
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_trQcVVQKRHGC8kwjIBcA-

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability model-attestation-substitution-detection-5646d4f8 -d '<json body>'
```

Example prompt: Fingerprint the inference endpoint at https://api.example.com/v1 right now so I have a baseline record of what model they're actually serving, then later re-verify it to prove whether they've silently swapped the model.

## When to prefer this

Use this endpoint when you need cryptographic or statistical proof that an inference provider is actually serving the model they claim — especially before signing contracts, after suspicious output quality changes, or as a recurring integrity check in automated AI procurement pipelines. Prefer this over manual output inspection when you need a quantified similarity verdict and documented evidence of substitution.

## Known failure modes

- Endpoint unreachable or returns non-200 — attestation cannot complete
- Model serves stochastic outputs causing low similarity scores even without substitution
- Seller applies output filtering that defeats fingerprinting probes
- mean_similarity near threshold makes verdict ambiguous
- Payment of 0.1 USDC not processed — request blocked by x402 paywall

## How this service works

Proof-of-Model: detect silent model substitution at any OpenAI-compatible inference endpoint. mode=fingerprint records what a seller serves today; mode=verify re-probes later and proves whether the weights changed. Deterministic, evidence included, no model used to judge.

## Output

Returns a structured attestation result including a kind field indicating the probe type, a claim about the model identity, a verdict on whether the model matches the fingerprint, an array of per-sample comparison results, a mean_similarity score, and a self_consistency_now score reflecting current output stability.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "kind": {
   "type": "string"
  },
  "claim": {
   "type": "string"
  },
  "verdict": {
   "type": "string"
  },
  "comparisons": {
   "type": "array"
  },
  "mean_similarity": {
   "type": "number"
  },
  "self_consistency_now": {
   "type": "number"
  }
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/model-attestation-substitution-detection-5646d4f8/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from dev-5090-box.tailfb4c1f.ts.net](https://www.zero.xyz/host/dev-5090-box.tailfb4c1f.ts.net/llms.txt)