destill.ai · Trust Center · Also: insurers · high-risk deployers · science publishers

Five more audiences, one kind of evidence.

Each section says the job in a sentence, what we can give you today, and where it stops. The status badge next to each item is read from one file, so it cannot say "built" on one page and "pilot" on another.

Advisers · Banks · Public bodies and universities · CISO and SOC · Creators and platforms

Advisers and multipliers

Your job: your clients ask how to show what their AI did, and you need something you can explain without becoming its vendor.

A receipt anyone can check

Show a client a signed receipt and let them verify it in their own browser. You do not need our code or our servers to explain the result.

Templates to start from

The Trust Center has a DPIA template and a threat model. One evidence format for all three modules is still on the roadmap .

Where it stops: we do not certify anyone, and a receipt does not make your client compliant.

Banks

Your job: AI agents prepare credit decisions and customer replies, and later you must show an auditor what the agent actually did.

Every agent step signed

Each step goes into a signed, time-stamped record. A later change to the record is visible.

Independent time stamp

The time comes from an outside RFC 3161 authority, not from us. Full checking of its certificate chain is still on the roadmap .

Keys in hardware

Keeping the signing key in an HSM is planned, not built. Ask us before you rely on key custody.

Public bodies and universities

Your job: you use AI for public tasks or research and must be able to account for it — to citizens, auditors or a research-integrity board.

Accountable agent work

Receipts can be checked offline and without us, which matters when the checker is a court of audit or an ethics board.

Human approval in the record

Today approval steps appear in the receipt. A separate signature by the reviewing person is planned.

Lab and instrument images

An image sealed at capture shows whether it changed since its seal. See the page for science publishers.

CISO and SOC

Your job: stop injected instructions before they reach your models and tools, and keep a record you can use in an incident review.

AEGIS screen

AEGIS screens prompts and tool inputs for prompt injection on a plain CPU, before the model sees them. Known gaps: role-play and story-style jailbreaks, false alarms on non-English text, paraphrased attacks.

No detection rate yet

We publish no detection rate until a reproducible public rerun exists. Test it against your own traffic in a pilot instead of trusting a number from us.

Finding in the receipt

Writing an AEGIS finding into the signed receipt is planned, so an incident review sees screen and agent steps in one record.

Our own threat model is public: threat model template and Trust Center.

Creators and platforms

Your job: find out which copy of your work leaked, and label synthetic media the way the AI Act asks.

Mark in images

An invisible, key-bound mark in each copy. Reading it shows which recipient the leaked copy belonged to .

Mark in video

Works in demos and pilots; not yet a production service.

C2PA bridge

Export and import of C2PA content credentials is planned, not built.

More for creators (German): Creator-Eigentum und Lizenzschutz.

Where it stops: a mark shows which copy leaked. It does not prove who made the work, and it does not show whether content was made with AI.

What we prove and what we do not

Verify a receipt yourselfRequest a pilot