P2 · Verifiable AI

AI Audit Log Proof

Hide the prompt, input data and model internals
Prove the output came from authorized instructions and inputs

Seal AI decision attribution with a ZK proof at decision time. Make past rationale recoverable after model updates.

Financial services · Insurance · Healthcare AI 6 min read
live in production since 2025 · Public-infrastructure PoC in production · ETHGlobal AI Agents 2026 Finalist
01 · THE PROBLEM

Three voices from the front line.

  • Internal audit / compliance

    “To reconstruct the basis of an AI decision later, we need a trail of the inputs, model and process”

  • Legal / executives

    “We can't explain AI-driven decisions to regulators or shareholders”

  • CISO / security

    “We can't detect tampering of AI decisions, so accountability is blurred”

02 · THE SHIFT

Hand over the source, or just the facts?

Change what reaches the AI, and the leakage risk goes with it.

Without Lemma
Hand over the original
prompt:
credit decision for case ○○
model:
gpt-internal-v4
params:
temp=0.2 …
response:
approved
timestamp:
2024-08-15…
↓ all of it goes to the AI / outside
With Lemma
Hand over just the facts
agent:
did:lemma:agent-credit-reviewer
modelId:
gpt-internal-v4
policyHash:
0x3d90…
inputCommitment:
0x7a2c…
outputCommitment:
approval = policy-compliant
satisfiesPolicy:
true
ZK verified:
✓ VALID
↓ only the necessary facts to the AI

At the moment the AI decides, the model used, the facts input, the criteria applied, and the final conclusion are fixed as one verifiable trail. The raw data stays in-house; what leaves is only the fact of "when, which model, on what basis, decided what." Past decisions stay immutable even as the model updates, and regulators, auditors and claimants can independently verify the same trail without disclosing the original data.

See the technical details ↗
03 · HOW TO CHOOSE

Choose on three criteria.

Only work that needs all three at once — pass without exposing, independent verification, tamper-proof — is Lemma's domain.

Method Pass without exposing Independent verification Tamper-proof
Access control only
Masking / anonymization
Encryption only
Logging / monitoring only
Lemma (ZK proof)the only one with all 3
04 · HOW IT WORKS

What's next

We enter through AI-governance and regulatory-readiness support and a PoC, and stay alongside you through to operations.

  1. A 30-minute review — identify the AI decisioning systems where accountability risk concentrates (lending, underwriting, clinical triage, public benefits).
  2. Narrow to 1–2 decisions (results) to prove — e.g. "a loan decline," "a credit-tier decision" — the facts sealed at decision time. Not the source data.
  3. Design connection and versioning — connection to your existing MLOps / inference pipeline, and version-fixing of the model identifier and applied guideline.
  4. Prove one decisioning system via a (quote-based) PoC — confirm decision-time attestation works on a single AI decisioning system.
  5. Hands-on support from rollout through operations — existing plan tiers (Civic / Critical / Compliance) serve only as a cost reference; the setup and pricing are designed together.

Tell us one AI decisioning system where accountability risk concentrates, in the first 30 minutes. No model implementation details or sensitive information required.

The bigger picture

The bigger picture this use case belongs to.

We map use scenarios across industries and workflows by the four axes.

See use scenarios for Verifiable AI in Solutions →

TRY LEMMA

Run it yourself.

No sales call needed — start hands-on with Lemma's products.