When your AI is questioned, have proof, not logs.
AI now denies loans, flags transactions, screens candidates, and recommends treatments. When those calls are challenged by regulators, auditors, customers, or courts, internal logs aren’t proof. They can be edited. Invoance seals the exact input, output, and model at the moment of the decision into a tamper-evident record anyone can verify, months or years later.
“▍
From AI output to independent proof
Your pipeline doesn’t change. After your model responds, one API call seals what went in and what came out, then anyone can re-check the math. Nothing here depends on trusting Invoance.
What gets sealed
Three things are fingerprinted and bound into a single signed record. Each is a SHA-256 hash, so the content never has to leave your control to be provable.
The exact prompt and context the model received, fingerprinted and never altered.
sha256(input) = 3f9a1c…e7b2The exact response the model produced, bound to the same record as its input.
sha256(output) = b7d266…a0c4Provider, model, version and key parameters, so the proof points to the exact model used.
model = gpt-4o-2024-11-20 · t=0.2Verification is recomputation, not trust
A third party doesn’t take our word for anything. They recompute the hashes and check the signature themselves. If one byte changed, verification fails.
The verifier hashes the input and output again and checks it matches the sealed fingerprint. One changed byte breaks the match.
The Ed25519 signature is verified against your organization’s published public key, proving the record came from you and wasn’t forged.
The append-only ledger entry shows the record existed at the stated time and hasn’t changed since it was sealed.
Every attestation produces a public link like invoance.com/proof/ai/att_b7d266a0. Open it from anywhere, with no login, no Invoance account, no access to your systems.
What attestation proves, and what it doesn’t
Attestation is a technical guarantee, not a legal or factual one. Being precise about the boundary is what makes the proof credible.
What you pay vs. what you cover
Slide to your monthly AI volume. See what attestation costs and how much unverified exposure it eliminates.
See what you pay and what you protect. Every event is anchored, timestamped, and independently verifiable, turning potential exposure into documented proof.
Built for the records regulators are about to require
High-risk AI systems are moving from “keep some logs” to “keep tamper-evident, auditable records.” An append-only signed ledger is exactly that substrate.
Art. 12 · high-risk loggingHigh-risk systems must automatically log events over their lifetime for post-market monitoring and audit. Enforcement for high-risk obligations begins Aug 2026.
AI management systemsCalls for traceability and records of AI system operation as part of a certified management system.
Medical devicesSoftware as a Medical Device using AI must support transparency and traceability of model decisions.
Financial regulationAI-driven credit decisions must be explainable, with the original decision preserved for adverse-action review.
Employment lawEmployers using AI in hiring may need to produce the original, unaltered AI outputs behind a decision.
Federal evidence rulesElectronic records are self-authenticating when generated and stored with verifiable process-integrity controls.
Your AI is deciding now. Start proving it.
AI attestation is available on all plans. No model changes required. One API call per output, permanent, tamper-evident, independently verifiable proof.
AI Attestation, Verifiable Proof of AI Outputs and Decisions
When your AI makes a decision, prove exactly what it said, when it said it. AI systems are making real decisions, approving loans, recommending treatments, screening candidates, drafting contracts. AI attestation creates a permanent, tamper-evident record of every AI input and output at the moment of generation. Designed for EU AI Act compliance, ISO 42001, and regulatory audits.