xBenchmark: Privacy-Preserving AI Model Evaluation

The Problem

AI model evaluation requires access to sensitive test datasets and model internals, but sharing either creates intellectual property and privacy risks.

Red-teaming and benchmarking require running sensitive prompts against models, but test datasets contain proprietary evaluation criteria, adversarial examples, and competitive intelligence. Sharing them with model providers defeats the purpose.

Model providers resist sharing weights or internal metrics for evaluation, creating a trust deadlock where neither party can verify the other’s claims.

The Old Way

The PRIVATE.ME Solution

xBenchmark uses xCompute to evaluate models on split data. Test datasets and model responses are XorIDA-split so neither the evaluator nor the model provider sees the other’s complete data.

Evaluation metrics are computed directly on XorIDA shares using xCompute’s Boolean circuit engine. XOR gates are free (zero communication); AND gates use Beaver triples. The result is a score that both parties can verify without either seeing the raw data.

All evaluation runs are recorded in an HMAC-chained audit trail with DID-signed attestations. Results are reproducible and tamper-evident.

The New Way

How It Works

xBenchmark orchestrates multi-party evaluation where test data and model outputs are XorIDA-split and scored via xCompute without reconstruction.

Use Cases

Integration

import { EvalSession } from '@private.me/xbenchmark';

const session = await EvalSession.create({
  evaluator: evaluatorDid,
  modelProvider: providerDid,
  metrics: ['accuracy', 'toxicity', 'bias'],
  threshold: { k: 2, n: 3 }
});
const result = await session.evaluate(testSuite);

Security Properties

Verifiable Data Protection

Every operation in this ACI produces a verifiable audit trail via xProve. HMAC-chained integrity proofs let auditors confirm that data was split, stored, and reconstructed correctly — without accessing the data itself.

Ready to deploy xBenchmark?

Talk to Ren, our AI sales engineer, or book a live demo with our team.

Ship Proofs, Not Source

xBenchmark generates cryptographic proofs of correct execution without exposing proprietary algorithms. Verify integrity using zero-knowledge proofs — no source code required.

Use Cases