AI Work Rules & Evidence Session
AI is already in your team's hands. The rules and the record usually aren't.
A paid, practical half-day work session that leaves your team with clear AI work rules, named human control points, and a signed, tamper-evident Evidence PDF — built on your real work, not a generic template. Not a webinar, not a lecture, not a free pilot.
A build session, not training
In most teams nobody has clearly defined where AI is allowed, where it is risky, what data must never go into an AI tool, who checks the output, and who stays responsible for the final decision. In one structured session your team builds exactly that — and seals it as evidence.
task → AI use → artefact → failure check → correction → evidence → signed seal
participants work on real but safe examples — no client, patient, financial or confidential data is used in the session
What your company leaves with
Each participant (3–7 per team) builds and receives their own record; the company gets the consolidated readout. Concrete artefacts, not notes.
Per participant — the 5 / 5 / 3 / 2 / 1 ruleset
- 5 allowed AI use cases for their real role
- 5 forbidden / restricted use cases — where the human decides alone
- 3 data-boundary rules — what may and may not go into an AI tool
- 2 human-gate rules — where AI output must be reviewed before use
- 1 hallucination-checking ritual, practised live
- A failure / correction log — proof control was actually exercised
Sealed & for the manager
- Sigillum-sealed Evidence PDF per participant (Ed25519)
- Manager mini-summary — a one-page readout of what was built, the boundaries, and the residual risk
- GO / FIX / HOLD recommendation for the next AI control step
- Company AI Control Mini-Passport (Tier 3) — a consolidated, sealed team-level trust profile
The seal is real cryptography — and you can check it yourself
Each evidence record is Ed25519-signed (the "Sigillum" seal) and tamper-evident: if a single byte of the sealed record changes, verification fails. The Evidence PDF is the human-readable certificate of that signed record, verifiable against it — not a logo or a watermark.
This is the same "recompute, don't trust" discipline we run in public on our live Regulatory Change Radar — where anyone can download a verifier and re-check a signed evidence record over HTTP, with no account. You can run it before we ever speak.
human-built rules · Ed25519-signed evidence · independently verifiable
What it maps to — stated precisely
Article 4 of the EU AI Act requires providers and deployers to ensure a sufficient level of AI literacy of their staff, taking into account their roles. The session produces documented evidence of exactly that:
Role-specific rules
The per-role 5 / 5 / 3 / 2 / 1 ruleset is documented, role-aware AI literacy — written against each person's actual work.
Failure + correction log
Evidence that staff didn't just declare competence but exercised it — caught a real AI failure and corrected it.
Sealed Evidence PDF
A signed, tamper-evident record you can show a partner, client or auditor — and they can re-verify.
Evidence, not opinion
Built on the same evidence/seal discipline we use on our own production AI systems.
Admissibility and legal sufficiency are always for a qualified lawyer and the relevant authority to decide.
Format & price
A practical working session — delivered in person or remotely by video. Deliverables are identical either way; the format affects only logistics.
We're early on purpose — you would be among our first paid sessions, at a founding rate. We'd rather earn your reference than borrow someone else's.
Book a 20-minute call
A short call to check whether your team has a real gap here, and to walk you through a sample of the exact deliverable format. Write one sentence about how your team uses AI today — we'll reply with two time slots.
Email [email protected]Margelis.ai · Rūpestėlis Holding UAB · Vilnius, Lithuania · EU