A claim photo that proves itself
Sealed at capture in the claims app, a signed receipt for every AI step: one chain from the photo to the decision letter, which you can break yourself in the demo.
destill.ai · All sectors · Deutsch · Trust Center
State of 5 November 2026
Each card says in one sentence what the product is for and takes you to where you can try it yourself. Where there is no public demo, the card says so. The measured limits of every product are on the overview for CIOs, exactly as they stand in the build record.
Everything here is measured in the lab (technology readiness level 6), not yet in a customer’s operation. Films exist for SEAL, TRACE, ProofCam, AEGIS and SIRENE, DOSSIER, RECEIPT, the developer page, CHAIN, CASE and VAULT, plus four market films: first a short version of about 30 seconds, below it the full film with the walkthrough of the real page. The voice in the films is synthetic (neural text-to-speech).
Why evidence anyone can check is worth more than trust: SEAL, TRACE, POAW and the firewall in one film, with a real receipt and a photo on the check page.
Source of the numbers in the film: seal 0 false alarms on untouched photos and 6 of 6 kinds of tampering (FORTRESS lab benchmark, small sample) · firewall about one millisecond per check (median 0.31 ms over 563 checks, destill-trust gateway measurement) · readiness level 6, no paying customer yet, patents filed as US provisional applications, not examined.
Sealed at capture in the claims app, a signed receipt for every AI step: one chain from the photo to the decision letter, which you can break yourself in the demo.
For journals: a submission rule instead of a detector race. The seal proves "unchanged since capture", never "not AI-generated".
A seal set at acquisition, carried inside the image file, with an open verifier. Inside the instrument it is a proposal for a partner, not built.
Lays an invisible, fragile seal over a photo or video at the moment of capture, only for a live capture from your own camera, never for an upload. If anything is changed afterwards, the seal breaks and the check shows where. It proves “unchanged since capture”, never “not AI-generated”; the seals are not signed today.
Burns an invisible per-recipient mark into every copy, in images, video and audio, and as an SDK also in JSON and CSV data. A leaked copy gives the key, your register the account. Photographing a screen, an AI redraw, a slight rotation and averaging two copies can defeat the mark.
The camera in your phone’s browser that seals every photo with SEAL as you shoot and gives it a proof code with a QR code. Uploaded pictures get no authenticity seal. A photographed screen is not reliably detected.
AEGIS screens every question for attacks before the model, on the processor; SIRENE checks every answer before it goes out and stops it when it gives away the system prompt. We publish an AEGIS detection rate only after a public rerun. SIRENE catches secrets copied word for word, reworded ones not yet.
Checks a credit file in your browser: ID and selfie sealed live, three signed steps of the credit agent, a decision letter with a recipient mark. If anything was altered, it names where the chain broke.
Seals a receipt photo at the moment it is taken in the app. The check later finds edits and the reuse of a receipt.
The SDK on one page, without an account or an API key: signing secret generated in your browser, code comparison, marking a whole archive, and the per-viewer player mark (STREAM).
Checks one claim package end to end in your browser: damage photo with seal, three signed AI steps, a notice with a recipient mark. Any manipulation names the break point.
The customer photographs the car in six guided views, each photo sealed at the moment it is taken. The insurer sees whether the photos are unchanged, complete, all different, and taken on one phone in one visit.
Turns a leak finding into one signed case file (location, time, mark, recipient, fetch log) plus a notice text under DSA Art. 16, verifiable offline.
Replays an AI agent run from its receipt and names the first step that deviates. The report is signed and verifies offline.
Shows which of your texts, code or data went into someone else’s material or AI training, as a report with locations and method.
Adds 7 metrics to the signed monthly report on a policyholder’s AI, each with definition and limit, checkable offline by an actuary.
Guards against your AI leaking system prompts, recipes or tariff rules: fixed rules in front of the model, plus an optional leak judge.
Marks a JSON or CSV dataset per recipient, so a leaked copy names the recipient’s key.
Puts a per-recipient mark into audio that survives MP3 and AAC. The shipped mark is audible.
Chooses the strength of the mark by the image content. Every report states the hit rate and false-hit bound of the read path that found the mark.
Four weeks, up to 100 of your files, one product of your choice. The success measure is agreed before the start.
Pages per sector: Banks · Insurers · Media and publishers · all sectors
Lebensfluss e.V., Gratkorn bei Graz, Österreich · ZVR-Zahl 1758759096 · Impressum · Datenschutz · Trust Center