EU AI Act · Annex III deadline 2 December 2027

The evidence file your AI system will be asked for.

TraceSeal classifies your AI system under the EU AI Act, collects tamper-evident logs of who did what with which model, and issues a signed Evidence Pack — PDF, JSON and a cryptographic seal — in minutes.

No account for the check. Not legal advice — a technical record your lawyer or auditor can work with.

Recruitment & HR tech·Credit & insurance·Education·Agencies
Evidence Pack · #1 SEALED
System
Talent Screening Assistant
HIGH-RISK · ANNEX III 4READINESS 71%
248
Events
31
Approvals
28
Controls
Pack hash (SHA-256)
9f2c7a1e4b6d8c0a3f5e7b9d1c2a4e6f
8b0d2c4a6e8f0b1d3c5a7e9f2b4d6c8a
Ed25519 signaturevalid
Issued by Northwind Recruitment Ltd2026-11-16

Three steps, one afternoon

01
Classify

Twelve questions mapped to Article 5, Annex III and Article 50. You get the level, the reasons and the exact articles that apply.

02
Connect your logs

One webhook. Point OpenAI, Azure OpenAI, Ollama or your own app at it. Every call is normalised to who, what, which model, which data, who approved.

03
Issue the pack

One click builds the PDF, the JSON and the seal. The hash chain proves nothing was changed after the fact.

What is inside the pack

An auditor does not want a dashboard screenshot. They want a fixed record with a date, a scope and a way to check it was not edited. That is what the pack is.

  • ✓Classification: level, reasons and Annex III reference
  • ✓Readiness per standard, with the open items named
  • ✓Human oversight: who approved what, when, with a signature
  • ✓Log summary: events, models, data categories, period
  • ✓Integrity: hash chain head, Merkle root, pack hash, Ed25519 signature
  • ✓Verification page and QR code for the auditor
Event record (normalised)
{
  "seq": 128,
  "source": "openai",
  "type": "inference",
  "actor": "recruiter@northwind.example",
  "model": "gpt-4.1-mini",
  "input_summary": "Rank 42 applicants for warehouse role…",
  "data_categories": ["email", "national_id"],
  "occurred_at": "2026-11-16T09:12:44.001Z",
  "approval_id": "ap_7f3c…",
  "prev_hash": "3c5a7e9f…",
  "hash": "9f2c7a1e…"
}

Personal data is masked before it is stored. The hash covers the record, the previous hash and the tenant, so a deleted or edited row is detectable.

Mapped to the standards being written now

The harmonised standards for the AI Act are in development. The checklist ships as a draft mapping, versioned in every pack, so the file does not go stale.

draft mapping · 2026-11
StandardScopeCoversControls
prEN 18228AI risk managementRisk register, evaluation, treatment, residual risk7
prEN 18229-1Logging and record keepingEvent scope, content, integrity, retention, export7
EN 18284Data governanceSources, quality, provenance, representativeness7
EN 18283Bias and fairnessAffected groups, metrics, mitigation, monitoring7

Pricing

No per-client hours. The pack costs the same whether it takes you ten minutes or ten days.

Risk check
€0forever

Classify one AI system, get the gap list and the articles that apply.

  • ✓Annex III classification
  • ✓Gap list with next steps
  • ✓No account needed
Run the check
Evidence pack
most bought
€490one-off

One system, one signed pack. For SMBs that need the file, not a subscription.

  • ✓PDF + JSON + Ed25519 seal
  • ✓90 days of logs
  • ✓Public verification page
  • ✓Checklist across 4 standards
Start
Monitoring
€149per month

Keep the evidence current: logs, monthly pack, oversight records.

  • ✓Up to 3 systems
  • ✓Unlimited events
  • ✓Monthly signed pack
  • ✓Approval trail
Start
Agency
€999per month

For consultancies that deliver AI Act files for several clients.

  • ✓Up to 20 systems
  • ✓White-label PDF
  • ✓Client workspaces
  • ✓Priority support
Talk to us

Prices exclude VAT. Payment by card. The free risk check never expires and needs no account.

Questions people actually ask

+Is this legal advice?

No. TraceSeal produces a technical record: what your system does, what you logged, who approved it, and whether the file was changed. Legal qualification stays with you and your lawyer.

+We only use ChatGPT through the API. Is that enough?

Yes. Point your API calls at the webhook once and every request is recorded with the model, the data categories and the user. You can also import an export.

+What if the standards change?

Every pack states the checklist version and the standard drafts it used. The catalogue is data, not code — updating it does not invalidate old packs, it creates a new version.

+Can the auditor verify the pack without us?

Yes. The PDF carries the pack hash, the public key fingerprint and a QR code to the public verification page. No login, no trust in us required.

+Do you store our prompts?

Only a masked summary: personal data (emails, phones, IBAN, national IDs) is replaced before storage. The full payload is not kept.

Find out what your system is, in two minutes.

Twelve questions. No account. You get the classification, the reasons and the three things to do next.

Start the free risk check