GovCopilot without leaks: AI practice in official work

This is not "attended and got inspired" — you leave with a working package for your department:

• The skill of rapid admissibility — determining in 2–3 minutes whether a task can be handed to AI: red, yellow and green data zones.
• A library of safe prompt templates for typical official tasks: a reply to a citizen, a summary, a memo, minutes, a presentation.
• A data admissibility checklist and a human review template — so that a model's draft does not go out without human review.
• A pilot use case passport for your unit: where the routine is, what data is involved, which model is admissible and what oversight is needed.
• A draft local regulation on "how to use generative AI in the unit" — ready for discussion with information security and management.

Guidelines for the workshop format; exact timelines and the composition of the training perimeter are fixed before the start:

Expected result: after 1 day — a working package of templates and a pilot passport for your department; after 2 weeks — the first scenario tested on a real task; after 1–2 months — a draft local regulation agreed with information security and management.

Each module closes with an artifact, not with notes.

1. Where AI is genuinely useful and where it creates risk. An overview of typical official work scenarios and an honest examination of the boundaries. Artifact — a map of "good and bad" scenarios.
2. Red, yellow and green data zones. A rapid admissibility map and anonymization rules: what can be sent to a model and what cannot.
3. Prompting for official tasks. Templates for replies, summaries, memos, minutes and presentations — with a focus on format and verifiability.
4. Practice on work cases. Citizen inquiries, a memo, minutes with assignments, an analytical note — each participant receives a package of real unit tasks.
5. Staff exercise "A manager's assignment in 45 minutes." Teamwork assessed on three criteria: quality, speed, compliance with information security constraints.
6. Implementation passport. Drafting one real scenario for your unit — a document you can discuss with management and information security.

The whole day is teamwork in groups of 4–5, where information security and the legal function are built into the practice rather than delivered as a separate lecture.

• 1 day (8 academic hours) or 2 half-days; groups of up to 30 participants.
• Teamwork in groups of 4–5 in a separate training perimeter.
• No AI background required — only your current official routine.
• No real personal data or restricted-access official information is loaded into the training perimeter: assignments are built on synthetic or anonymized materials.
• A Russian model stack (GigaChat, YandexGPT) as the primary perimeter; external models are demonstrated only in an isolated environment.

• Specialists and consultants of government bodies — to prepare documents and replies faster.
• Department heads and deputy managers — to standardize quality across the unit.
• Digital transformation officers, analysts and project offices — to launch a managed AI pilot.
• Representatives of information security and legal units — as co-moderators who set the boundaries of the perimeter.

We are an engineering and consulting collective that implements CRM, ERP, RPA and AI in real processes itself, including in the public sector and secure perimeters. That is why the workshop is built around what a public servant actually needs:

• Results through artifacts, not theory. Every block closes with a template, checklist or pilot passport that can be put to work.
• Security at the core, not as a boring lecture. Information security and legal constraints are built into the assignments themselves rather than presented as a separate block.
• Honest about limitations. We say where AI helps and where it creates risk; which scenarios require approval from information security and lawyers, and which should not be touched at all.
• We do not promise "N times faster." We promise a method and "before → after" verifiability: you are left with a measurable process, not a pretty picture.

Is this a lecture about neural networks? No. This is a staff workshop: you solve your own official tasks in a safe perimeter and leave with templates and a pilot passport, not with "inspiration."

Can Chinese models be used? In public external clouds for the public sector — no. Qwen and DeepSeek are demonstrated only locally or in a private isolated perimeter on the customer's infrastructure. This is one of the workshop's basic principles.

Who is responsible for the result? AI prepares the draft, while the decision and legally significant actions are taken by a human. Human review is built in as a mandatory step — this removes both the fear of "making a mistake" and the blurring of accountability.

Is any special data needed to participate? No. All practical assignments are built on synthetic and anonymized materials — no real data is loaded into the training perimeter.

Cohorts are formed on request: dates, cost, venue and the composition of the training perimeter (including the access model for information security units) are agreed individually.

First step: enroll or request a cohort for your team — we will send the terms, the curriculum and confirm a date for your unit.

If you are unsure: send us one of your typical tasks, and in a free short review we will show how it goes through the safe cycle and where the risk lies.

Not relevant to the course at all? Start with the free cheat sheet on data admissibility zones — what can and cannot be handed to AI in official work.

[Enroll in the next cohort] · [Request a free review of your task] · [Download the data zones cheat sheet]

Does the unit need to prepare before the start?

Yes, light preparation is needed: 1–2 days before the workshop we agree on the team composition and collect 3–5 typical tasks that the work will be based on. That way the training runs on your routine rather than on abstract examples.

How many people can study in one cohort?

A comfortable size is up to 30 participants, with working teams of 4–5 people. A separate closed cohort with an agreed access perimeter can be held for a unit.

What if external AI services are prohibited in the unit?

The workshop is built in an isolated training perimeter on Russian models (GigaChat, YandexGPT). External models are demonstrated only locally or in a private environment and are not used for real data.

Do I need to bring my own documents and data?

No. Assignments are built on synthetic and anonymized materials; no real personal data or restricted-access information is loaded into the perimeter.

How are the interests of information security and lawyers taken into account?

Information security and legal constraints are built into the assignments themselves. Information security representatives can participate as co-moderators and set the boundaries of the perimeter right away, and the draft local regulation goes for approval with their edits.

• Training and acceleration — if you have a task of raising AI competencies across an entire team.
• AI for the public sector — if you need not only the skill but also the implementation of a registry solution.
• Courses and training — the full catalog of programs.
• Blog on learning and competencies — how to train a team and not lose the result.