destill.ai · Deutsch · Trust Center
For anyone who runs, audits or insures AI agents
Who proves what your AI actually did?
An AI agent decides, something goes wrong, and the only witness is a log that can be changed. Proof of Agent Work turns every agent run into a receipt: changed means invalid, and anyone can check it with standard tools.
Try it yourselfRequest a pilot
Operators
You will be asked what the agent did. Today you answer from a database you control, which is exactly why it convinces nobody.
Auditors
You want to check a trail without trusting the system that wrote it.
Insurers
You price what you cannot see. A signed run shows which controls actually ran.
The film
The opening scene, as a short clip.
The full explainer
Chapters
Pick your sector
Insurers
Six roles, sealed claim photos, the claim receipt.
Banks
Credit decisions, internal audit, model risk.
Healthcare and pharma
QC steps, audit trails, sealed images.
Pharma Medical Affairs
Medical enquiries, a receipt per answer.
Vehicle assessors
Sealed damage photos for reports and claims.
Car rental and carsharing
Sealed photos at handover and return.
Creators
Show what is yours: an invisible mark at upload.
Try it yourself: a signed receipt you can check
This is the kind of receipt you would get from every run of your agent after the pilot. Synthetic data, real cryptography: an ML-DSA-65 signature and an RFC 3161 time stamp from an independent authority. Check it in your browser, then check a copy where one value in step 3 was changed.
Claim payout by an AI claims agent (fictional)
| Step | What the agent did | Time (UTC) |
|---|---|---|
| 1 | Claim receivedclaimId: CLM-2026-004817 · line: motor · channel: customer portal · documents: 3 × hash | 09:14:02 |
| 2 | Policy retrievedpolicyId: POL-88-2231-7 · cover: comprehensive · deductibleEUR: 500 · inForce: true | 09:14:04 |
| 3 | Damage estimatedmodel: vision-estimator 3.2 · estimateEUR: 4700 · confidence: 0.91 | 09:14:19 |
| 4 | Fraud screenedscore: 0.07 · threshold: 0.35 · result: clear | 09:14:21 |
| 5 | Payout proposedamountEUR: 4200 · basis: estimate minus deductible | 09:14:22 |
| 6 | Human approvalrole: senior adjuster · approver: hash | 09:31:47 |
| 7 | Payment instructedamountEUR: 4200 · payee: hash | 09:31:49 |
Check this receiptCheck a tampered copy
Download the receipt (JSON) · Or drag the downloaded file into the verifier yourself, edit one number in a text editor, and drag it in again.
Credit decision by an AI credit agent (fictional applicant)
| Step | What the agent did | Time (UTC) |
|---|---|---|
| 1 | Application receivedapplicationId: APP-2026-01187 · product: consumer loan · amountEUR: 18000 · termMonths: 60 · documents: 2 × hash | 08:02:11 |
| 2 | Credit bureau report retrievedsource: credit bureau · report: hash | 08:02:13 |
| 3 | Score computedmodel: credit-scorer 2.4 · score: 612 · cutoff: 640 | 08:02:14 |
| 4 | Affordability checkeddebtToIncome: 0.41 · limit: 0.4 · result: above limit | 08:02:14 |
| 5 | Decision proposeddecision: decline · reasons: score below cutoff, debt-to-income above limit | 08:02:15 |
| 6 | Human reviewrole: credit officer · reviewer: hash · outcome: decline confirmed | 09:40:02 |
| 7 | Customer notifiedchannel: letter · explanationOffered: true · letter: hash | 09:40:05 |
Check this receiptCheck a tampered copy
Download the receipt (JSON) · Or drag the downloaded file into the verifier yourself, edit one number in a text editor, and drag it in again.
Visual quality-control step (fictional batch)
| Step | What the agent did | Time (UTC) |
|---|---|---|
| 1 | Batch receivedbatchId: B-26-0417 · product: tablet 50 mg (fictional) · line: L3 | 06:15:00 |
| 2 | Images capturedstation: inspection camera 2 · images: 3 × hash | 06:15:42 |
| 3 | Defects classifiedmodel: vision-qc 1.7 · defectsFound: 2 · classes: chipped edge, spot · confidence: 0.94 | 06:15:44 |
| 4 | SOP limit checkedsop: SOP-QC-112 v5 · acceptLimit: 3 · result: within limit | 06:15:44 |
| 5 | Disposition proposeddisposition: forward to QA review | 06:15:45 |
| 6 | Human approvalrole: QA reviewer · approver: hash | 07:02:19 |
| 7 | Record archivedsystem: eQMS · record: hash | 07:02:21 |
Check this receiptCheck a tampered copy
Download the receipt (JSON) · Or drag the downloaded file into the verifier yourself, edit one number in a text editor, and drag it in again.
Your engineers can repeat the check offline with OpenSSL 3.5 or later and the published script: check pack.
What you get
Each item carries its status from our feature register: Built, Pilot or Roadmap. Hover a label for the evidence behind it.
One signed receipt per agent run: every step, in order, with model and version. Post-quantum signature ML-DSA-65 (FIPS 204). Time stamp from an independent RFC 3161 authority; nobody can backdate the receipt, us included. The human approval signed with the reviewer’s own key. Append-only transparency log with inclusion proofs. Keys in hardware (HSM).
The legal frame
Where a receipt helps you evidence an obligation. It does not fulfil the obligation for you and does not make anyone conform with the law by itself. Not legal advice.
AI Act Art. 12, 19 and 26(6)
High-risk AI logs events automatically; logs are kept for at least six months. The receipt helps you evidence that a log was not changed since it was stamped.
Source: eur-lex.europa.eu
AI Act Art. 86 and GDPR Art. 15(1)(h), 22
Right to an explanation of the role of the AI system, to meaningful information about the logic, and to contest. The signed trail helps you answer with the steps that actually ran; the explanation itself you give.
Source: eur-lex.europa.eu
Product Liability Directive (EU) 2024/2853, Art. 9 and 10
Disclosure of evidence and presumptions of defect for software. A receipt helps the party holding it evidence what the system did; no outcome is guaranteed.
Source: eur-lex.europa.eu
The honest limit
A receipt proves who signed which steps, and when. It does not prove that the agent was right, and it does not prove that nothing was left out: it covers what was logged. The SEAL shows a change made after capture; it never says a picture is "not AI-generated". Keys are held in software today; hardware custody is on the roadmap.
The pilot: 8 weeks, price after scoping
Scope
One agent workflow in your environment. POAW complements your logging and trace tools rather than replacing them. Verification in the browser or offline. Final report with go/no-go.
Price frame: Price after scoping, based on your process and environment.
Success criteria, agreed in writing at kick-off
- At least 99 % of AI steps in the pilot workflow produce a verifiable receipt.
- Your team verifies a receipt on its own, without our help (OpenSSL).
- Extra effort per work step is measured; we set the target together at kick-off.
Source: our written pilot offer; the targets are what we commit to measure, not results from an earlier pilot.
What you provide
- One concrete workflow and an owner
- Test data or test images and access to a test environment
- Contacts from IT and compliance or audit
- Internet access when signing (for the time stamp)
- Your own legal assessment
- Connection to your agent
Want the written pilot offer? Choose "Talk about a pilot" below and we send it with our reply.