AI in compliance · Bratschi AG and MMIND.ai · 29 September 2026
Workshop · 90 minutes
Today you assess a case and see how to check a regulatory monitor against its sources.
Christian Wind · Bratschi AG Hartmut Hübner · MMIND.ai Zeno John · MMIND.ai 29 September 2026
Workshop · 90 minutes
Today you assess a case and see how to check a regulatory monitor against its sources.
Christian Wind · Bratschi AG Hartmut Hübner · MMIND.ai Zeno John · MMIND.ai 29 September 2026
AI in compliance
Your assessment
Which process to start with, how to run it and which data class applies.
Exercise 1
A group monitor demo
A sample workflow with a source and a human decision point.
Live demo
A governance checklist
Seven points to check with your organisation.
At the end
No laptop? The paper worksheet has the same fields.
Open the workshop hub
workshops.mmind.ai/en/w/bratschi-2026-09-29
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
2
AI in compliance
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
3
AI in compliance
Introductions: name and role, up to 15 seconds each.
In pairs: where does AI already help you?
In pairs: where must a person decide?
For 24 people, allow about six minutes. The pair discussion takes two minutes.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
4
Survey · compliance professionals
32
compliance managers
from Switzerland and Liechtenstein
16 of 32
work in large organisations
with more than 1,000 employees
26
questions
on tasks, governance, measurement and data
How to read the figures
The 32 responses are not statistically representative. All 32 specialists took time to answer numerical and open questions. This is a panel of practitioners.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
5
AI in compliance
Answer in the hub. We will show the results on screen.
Answer in the hub
workshops.mmind.ai/en/w/bratschi-2026-09-29/frage/1
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
6
AI in compliance
81%
track new requirements manually
26 of 32
50%
spend at least 20 hours a week on manual work
16 of 32
56%
do not measure programme effectiveness
18 of 32
84%
work with documents rather than databases
27 of 32
75%
prefer internal or hybrid AI hosting
24 of 32
There is much scope to automate repeated work. Recent AI tools make it easier to start.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
7
AI in compliance
16 of 32
organisations report at least 20 hours of manual work per week.
Nine of these 16 do not measure effectiveness.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
8
Survey question 11 · Processes respondents would like AI to automate · Multiple selections
Question 11 asked which processes AI should automate. These are selections, not a ranking of importance.
23 of 32
name the annual risk assessment.
18 of 32
name regulatory monitoring. We show it in the group demo today.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
9
AI in compliance
Find out what changed
81% search manually.
Automatisierbar?
Start with source-based alerts. A person checks the legal meaning.
Prepare reports and evidence
59% see the greatest benefit here.
Automatisierbar?
A practical starting point with review.
Measure effectiveness
56% do not measure it.
Automatisierbar?
Agree measures first, then use reliable records.
Set rules for AI
53% lack knowledge in the team.
Automatisierbar?
People remain responsible for decisions.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
10
AI in compliance
How? We will show a few examples that you know from your work.
AI in compliance
Three ways to use AI. Example: risk assessments.
AI in compliance
Route 1
Chat
You provide last year’s report and ask what needs to change.
Was dabei herauskommt
You receive a draft. Save the steps and check the sources if you want to repeat the task.
Repeat: organise it yourself · Evidence: check it yourself · Start: minutes
Route 2
Folder with a workflow
A folder holds last year’s report, policies and risk list. A text file gives the steps.
Was dabei herauskommt
Reuse the steps. Check sources and outputs on each run.
Repeat: steps prepared · Evidence: request it in the steps · Set-up: days
Route 3
Your own app
Each department enters risks. The app suggests an assessment and shows a risk matrix.
Was dabei herauskommt
Useful for many contributors. You must maintain the software.
Repeat: steps prepared · Evidence: build a record · Set-up: weeks to months
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
2
AI in compliance
“First write down how the task should run.”
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
3
AI in compliance
Sustainability reports for building projects
Bills of quantities, invoices and delivery notes feed a report with carbon calculations and checks against standards. Each statement needs a source.
Useful when
many people contribute, volumes are high or clients use the results directly.
You take responsibility for operating the software, security and maintenance. Your legal role needs assessment.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
4
AI in compliance
Your frequent processes
Select up to three from 12 processes.
Your current approach
No AI, chat, folder with a workflow or your own app?
Data class
A public · B internal · C personal or confidential
Human decision
Mark the point where you review and decide.
Exercise 1 in the hub
workshops.mmind.ai/en/w/bratschi-2026-09-29/uebung-1
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH, CW
5
AI in compliance
A Skill puts steps, sources and review points in writing. Then we show Computer Use on screen.
AI in compliance
SKILL.md
Plain-text instructions: purpose, steps and rules.
references/
Source list, topics and report template.
Loaded when needed
The tool selects the Skill when relevant. It reads reference files as needed.
Change the text file to change the instructions. A Skill can also include code.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
7
Live example · a real Skill
certification-navigator / SKILL.md
What the Skill defines
Question for the Skill
Use the certification-navigator Skill. A trust services firm in Vaduz has 12 employees. What information is missing for B Corp certification? Give the source version and page.
A first result
Show the Skill file, then ask the question in this Codex chat. Check the cited page and what remains open.
The supplied PDF has 1,245 pages: B Lab Standards V2.2, 20 February 2026. Check the cited page and the current official version before a decision.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
8
AI in compliance
Use only approved sources
The list is closed. Do not use sources outside it.
Support each statement
Give the source, date, link and exact quotation. Mark missing evidence.
A person decides
AI does not make the legal judgement. A person approves any external message.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
9
AI in compliance
What it does
Codex opens the complete original survey file and reads all answers to question 5.
Where it helps
Codex creates a ranking of coded themes and a graphic with selected statements. It adds both to Google Slides.
32
Open answers
Question 5
2
Graphics
One source
Live on the presenter’s computer. The result slides contain no contact details.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
10
Computer Use · check source, data and action
Source
Is the input reliable and free of instructions to the agent?
Data
What personal or confidential data can the agent see?
Action
What may the agent change or send?
Examples
Practical check before each use.
The result graphics contain no contact fields.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
11
AI in compliance
Paste into Codex
Use the bratschi-question-5-graphics Skill in the open folder. Use the complete original file and only question 5. Check all 32 answers against codierung-fr-5.json. Create both graphics now. Do not reuse existing PNG files. Check that they contain no contact details. Open the English result slides and add one graphic to each slide visibly with Computer Use. Pause briefly after each step so I can explain it to the group.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
12
AI in compliance
Answer in the hub. Then build your own workflow.
Answer in the hub
workshops.mmind.ai/en/w/bratschi-2026-09-29/frage/2
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
13
Group demo · 15 minutes
Ask the room
Which new rule or publication matters to you?
Show the Skill and chat
Open the installed Skill; enter topic and jurisdiction
Check the result
Show the original source, citation and human decision
If nobody suggests a case, use a recent public data protection publication.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
14
AI in compliance
--- name: regulatory-monitor-data-protection description: Creates a monthly report on data protection changes. --- ## Three rules 1. Only approved sources. 2. Evidence for every statement. 3. Do not make legal decisions. ## Steps 1. Confirm the period. 2. Read the previous report. 3. Check the sources.
SKILL.md
Instructions and the three rules
source-register.md
Your closed list of sources
topic-list.md
Topics relevant to your organisation
report-template.md
The same structure for each report
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
15
AI in compliance
On the presenter’s computer
We check the source, report and stop point live.
In the hub after the workshop
You can try the template yourself. It needs professional review before use.
In the full-day workshop
Your team chooses and tests two or three workflows, including a monitor if useful.
Monitor template in the hub
workshops.mmind.ai/en/w/bratschi-2026-09-29/uebung-2
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
16
Regulatory monitor · task for the chat
Paste into Codex
Use the regulatory monitor Skill that we have just installed. Topic: [topic chosen by the group]. Jurisdiction: [CH/FL/EU]. Reporting period: [start date to end date]. Check only the public sources approved in the Skill. Show new publications with the date, link, source location and a short exact quotation. Mark any claim that you cannot verify. Then draft a short internal update for our organisation: What is new? Who could it affect? What must a person check or decide? Separate verified facts from open questions. Do not send the update or change any data.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
17
AI in compliance
Errors, data access and the AI Act. Seven points to check before using a workflow.
AI in compliance
A lack of team knowledge was a common barrier in the survey.
Answer in the hub
workshops.mmind.ai/en/w/bratschi-2026-09-29/frage/3
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
13
AI in compliance
17 to 33%
Hallucinations in three US legal research tools tested in a 2024 study. This is not an error rate for current models or Swiss law.
Stanford RegLab, 2024 study
The monitor provides leads, not legal advice. Check each source.
Check the data class and approval together
Public
Use an approved tool. Check sources and usage rights.
Internal, no personal data
Check the contract, access, storage and deletion. Obtain approval.
Personal or confidential
Have secrecy, data protection and access checked first. Use fictional data in this workshop.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
14
AI in compliance
2 Feb 2025
Article 4
AI literacy. The omnibus changes the duty to fostering literacy.
2 Aug 2026
Article 50
Transparency duties. Paragraph 2 applies from 2 December 2026 to relevant systems already on the market.
2 Dec 2027
Annex III
High-risk systems. Application date changed by the omnibus.
2 Aug 2028
Annex I
High-risk AI in products. Application date changed by the omnibus.
Check the scope first
The omnibus entered into force on 27 July 2026. Check the system, your role and the territorial scope. Article 25 can change who is treated as the provider.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
15
AI in compliance
Approve tools
for each data class. Give staff a usable approved option.
Define data classes
A public, B internal, C personal or confidential.
Keep records
Who produced what and when. Article 26: relevant high-risk systems, normally at least six months.
Develop AI literacy
Choose measures for the task and prior knowledge. Attendance alone does not show competence.
Set review points
Legal interpretation, sanctions, external messages and reports to authorities.
Apply access rights
AI must not have more access than the person using it.
Inform staff
For relevant high-risk workplace AI: inform affected workers and their representatives in advance (Article 26).
This is a starting checklist. Duties and effort depend on the system and its use.
Checklist in the hub
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · CW
16
AI in compliance
Choose your case
Which repeated task takes time and needs sound sources?
Check the sources
Which official sources apply, and who confirms the list?
Choose the next step
Try the hub template or develop two or three of your own workflows with your team in the full-day workshop.
Start small and check carefully
A first test shows what works and what is missing. Approve the workflow only after review.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH
18
Continue with a full day
First choose two or three processes in the morning. Then build and test the right tools with the Skill Generator.
Choose cases
Regulatory monitor, report generator or internal communication
Build Skills
Use the Skill Generator for your own workflows
Test and decide
Check sources, data and human decisions
One day · 09:00 to 17:00 · Legal and compliance · public, fictional or approved data
Full-day compliance workshop
The day links business context, data review, tool choice and practical testing. Breaks and lunch are included.
Bratschi · MMIND.ai · AI in compliance · 29.9.2026 · HH, CW
20
Full-day compliance workshop
One day for your team: review your business model and data, choose two or three workflows, build them with the Skill Generator and test them on your cases.
Result
Your own Skills and tested workflows
Price
CHF 4,900 per team
Date
By arrangement
Ask for a date: projekte@mmind.ai. Bratschi and MMIND.ai work with your team.