WITH CHRISTINA HOFFMANN
Pioneer Circle.
You don’t have to use AI to be affected by it.
Help advance SMGI, the governance counterpart to AI I’m developing. Contribute funding, relevant expertise or the right introductions.
Propose a contributionA non-binding enquiry. No payment today.
THE PURPOSE OF SMGI
What SMGI is.
Societal Meaning Governance Intelligence
SMGI is a software architecture I’m developing to oversee how AI agents make decisions and act together. Its purpose is to recognise when a developing course of action is becoming unsafe or exceeding its authority, and intervene before the next consequential action.
Behavioural profiling is a core measurement element. It is intended to build a continuously updated picture of how decisions change within a task: how uncertainty is handled, how agents influence one another, and how pressure to complete an objective affects their choices. This concerns individual agents and the behaviour that develops between them.
A simple example
One agent reports a possible problem. Another summarises it but drops the uncertainty. A third treats that summary as a confirmed fact and prepares an action nobody authorised. Pressure to finish makes the sequence move faster.
SMGI is being designed to recognise these developing changes and connect them to intervention, for example holding the affected action while its justification is checked.
The architecture is being developed. Implementation and testing must establish what it can detect, how reliably it can intervene, and where its limits lie.
WITH CHRISTINA HOFFMANN / 1 MIN 40 SEC
Why this work.
Why now.
Building and testing a system takes time. So does finding people who can carry the responsibility.
I want to prepare before a serious failure shows us what was missing. In this short introduction, I explain why I am bringing people together now.
Read the transcript
You don’t have to use AI to live the consequences of decisions made by it. Consider a confidential report. One agent analyzes it, another prepares a summary, a third sends it outside the organization. Each step can look routine. Together, they can expose information that was never authorized to leave. Checking each action separately can miss the risks created by how those actions interact.
That is the problem I’m working on. For nine years, I’ve worked on human behavior, decision patterns, and what changes under pressure. That foundation now informs SMGI, the governance counterpart to AI I’m developing.
SMGI is being developed to assess how agents’ decisions and actions influence one another with the aim of intervening before harmful consequences spread. The system still needs to be built and tested. A patent filing is not proof that it works.
So why now? Because finding the right people, developing the system, and testing it properly takes time. I don’t want us to start preparing only after a serious failure has shown us what was missing.
I filed the U.S. provisional patent application. Now I’m moving the patent work forward and bringing together people who can help take the next steps. That is why I’m creating the Pioneer Circle.
You can support the patent work financially or contribute relevant expertise and introductions. Through selected updates and sessions, you can get closer to the thinking behind the work, examine its limits, and help work through the open questions. Confidential material requires an NDA.
Read the plan below. If this is work you want to help move forward, tell me what you could contribute.
WHY SMGI, WHEN AI SAFETY WORK ALREADY EXISTS?
Being able to act
is not permission to act.
Existing safeguards matter. The question is what they can establish when agents’ actions combine, context changes and consequences unfold faster than people can intervene.
01 Another AI checking the AI
Monitoring can catch mistakes. But models can share missing context or mistaken assumptions. Agreement between them does not itself establish safety or authority.
02 Deterministic rules
Rules can enforce boundaries. Correct enforcement cannot repair a specification that misses a dependency or relies on outdated information.
03 Human review
People remain responsible, but cannot prevent a consequence that unfolds before they can recognise it and intervene. Audit trails are not a substitute for timely control.
SMGI’s intended starting point is human-authorised standards, separate from the AI pursuing the objective. Its effectiveness, including missed risks and unnecessary stops, must be established through testing. This is not a claim of demonstrated superiority.
Read the reasoning behind this approach
My earlier articles set out the distinction between observation, rule enforcement and authority: The Monitor Is Inside the Box and Why Most AI Governance Solutions Will Fail As Control Systems.
AI control research has demonstrated improvements in bounded programming experiments. That supports those tested protocols, not a universal safety guarantee: Greenblatt and colleagues, AI Control. Deterministic authorisation can include context and relationships: Cedar’s authorisation model. Neither source is a validation of SMGI.
Background: International AI Safety Report 2026, multi-agent reliability risks. This research describes the problem, not validation of SMGI.
THE FOUNDATION
Nine years of work.
A concrete foundation.
I left a corporate CMO career to develop profiling and recommendation technology. That interdisciplinary work on human behaviour, decisions under pressure and interacting systems now informs SMGI’s measurement architecture.
ZHAW + Innosuisse
The foundational research involved ZHAW with Innosuisse support. ZHAW researchers published the related Personality ASPECTS model in the peer-reviewed Journal of Consumer Behaviour in November 2025.
600,000+ records
Across eight years of corporate client projects in automotive, finance and commerce. Over 400,000 records were sufficiently usable for applied profiling.
Applied technology
A proprietary recommendation approach linked three deliberate choices to motive and sensory preference signals, matching people with relevant shoes without a large personal behaviour history.
This foundation does not by itself validate SMGI’s application to AI. It concerns observable decision patterns, not a claim that AI has human personality or consciousness.
Research funding, ownership and background
Innosuisse support included up to CHF 1 million allocated to eligible third party research services in the ZHAW collaboration. This did not fund my own founder costs.
The existing proprietary commercial profiling and ASPECTS IP belongs to my company, Brandmind GmbH, now operating as NOMAMIND. The ZHAW publication is a separate research output. I hold the SMGI patent application personally.
Founder background and applied work / Personality ASPECTS research record
The longer-term plan: NOMAMIND and EXIDION
NOMAMIND: build and test
Applied research into how agents’ decision pathways drift, implementation and testing. The people building SMGI will work through NOMAMIND.
NOMAMIND’s research directionEXIDION: foundational research
A separate Swiss association building a network for research into learning, meaning, maturity, agency and responsibility under pressure, including the provisional Human Drift question. This public-benefit work may continue for decades; it is not an already funded lab.
EXIDION’s research agendaThe Circle can connect supporters to both tracks. Its monthly contribution supports my patent work. Funding for EXIDION is agreed separately with the association; Circle participation does not confer association membership or exclusive research rights.
THE MISSION AND NEXT STEPS
From groundwork
to a working system.
This is a mission, not a conventional startup or a quick tool launch.
- 01 / FOUNDATION
Research and
applied workThe profiling foundation informs SMGI’s measurement architecture.
- 02 / FILED
13 October
2026The U.S. provisional is filed. Patent pending, not a granted patent or proof that the system works.
- 03 / NEXT FILINGS
Advance the
patent work- U.S. Non-Provisional
- US$1,000 to $1,200
- European application
- Just under €5,000
- 04 / IN PARALLEL
Find the
right peopleSystems architects, engineers and technical reviewers to challenge, build and test the work.
- 05 / BUILD AND TEST
A first version.
Responsible pilots.Establish limits, then work with early customers prepared to share responsibility for bounded pilots. Later, aligned investors may help take tested work further.
These filing amounts come from the cost input I have received. They are not the budget for building the system.
Decision integrity
under pressure.
I need people who disclose mistakes, challenge assumptions and hold their standards under public and political scrutiny, even when it costs money, a partnership or an opportunity.
INSIDE THE CIRCLE
Stay close to the work
while it is being built.
Planned launch: 16 November 2026. Participation is within an agreed scope.
A monthly session with me
Up to 60 minutes, with questions and preparation beforehand. Attendance is optional; cancelled sessions are rescheduled.
Selected development updates
What has progressed, the reasoning behind it, its limitations and questions still open. Not unrestricted access to the architecture.
Real incident analysis
Examine a documented AI incident: what happened, which human patterns recur and where control failed. Work towards a useful requirement or test question. Complex cases may need several sessions.
This is a working circle, not a general discussion group, course or individual advisory service. Expertise contributors participate selectively. Confidential discussion requires the signed NDA.
Technical implementation, not general systems discussion.
SMGI already has a conceptual architecture. The immediate challenge is turning that foundation into a secure, observable and testable technical system.
People who have built systems
Software and systems architecture for complex, distributed or multi-agent systems. Backend and infrastructure engineering. Security architecture, permissions, sandboxing and controlled system actions.
People who can make them reliable
AI agent integration and technical evaluation. Reliability, verification and testing of safety-critical workflows. Observable measurement systems. Relevant introductions to people with this experience.
Systems thinking is valuable. It is not a contribution on its own. A contribution must connect to a defined work package, a relevant introduction or financial support.
You do not need confidential access to tell me what you have built, researched or implemented before. If there is a credible fit, we define a useful first contribution before any deeper access is considered.
THREE WAYS TO CONTRIBUTE
Help move the work forward.
Support the patent work
US$250/month
Fund the next patent steps. The contribution supports patent work, not a fee per session.
No fixed minimum term. Cancel with one calendar month’s notice. No charge before your agreed start. Any applicable tax and exact recurring total are confirmed before agreement or payment.
Contribute expertise
No fee, by agreement
Bring relevant experience. We assess fit and define a useful first task together, including scope, time commitment and result.
Selective participation, not an offer of employment. Implementation work requires a separate agreement.
Make a relevant introduction
No fee, by agreement
Connect me with potential builders, technical reviewers, patrons or, later, pilot customers.
No introduction quota or retrospective fee. General curiosity or available time alone is not a contribution.
What could a first task look like?
- Map an incident: trace a documented sequence, dependencies and missing information. Produce a concise failure analysis.
- Review a requirement: examine assumptions, integration constraints and edge cases in an agreed brief.
- Shape a test case: identify evidence that would distinguish an effective intervention from a missed risk or unnecessary stop.
These are possible starting points, not assigned roles. You do not need to understand the full architecture before getting in touch.
Support, not investment. No equity, ownership, licence, financial return or guaranteed patent or development outcome. Financial support does not expand confidential access.
Propose a contributionYOUR NEXT STEP
Where could you
contribute?
You don’t need a finished work package. Tell me what you have worked on, which part of the challenge you understand and what you could realistically contribute.
Keep this enquiry non-confidential. Submitting does not automatically send an NDA or grant access.
No payment today. No automatic subscription.
This enquiry creates no membership or payment obligation and does not subscribe you to marketing emails.
How your enquiry is handled
Christina Hoffmann, Hofibachstrasse 3, 8909 Zwillikon, Switzerland, uses your details to answer this enquiry and assess potential participation. Contact: christina.hoffmann@nomamind.com.
Your submission is stored in this website’s form system and sent by email. Website and email service providers process it to deliver these functions. No confidential material or payment information is requested.
You may request access, correction or deletion by email, subject to applicable retention obligations. Separate participation, confidentiality and any applicable consumer information will be provided before you enter into an agreement.
A FEW IMPORTANT DETAILS
Before you commit.
What can be shared, and why is an NDA needed?
You do not need an NDA to understand the public explanation or discuss whether your experience could be relevant. After we agree there is a fit, you receive the NDA before selected confidential material or discussion. Both parties must sign it. Even an analysis of a public AI incident may reveal non-public concepts; public information itself remains public.
The NDA does not unlock the entire architecture. Access remains limited to selected overviews, explanations and material needed for the agreed scope. The full provisional application, source code, internal calculations, thresholds, implementation parameters and unpublished European patent development are not included. Later filings do not automatically expand access.
When do we start, and what if I miss a session?
The planned launch is 16 November 2026. Individual start dates are confirmed before agreement. Sessions take place monthly on Google Meet, for up to 60 minutes. Attendance is optional. If I cancel a session, it will be rescheduled. Send questions beforehand to christina.hoffmann@nomamind.com or via Instagram.
How does cancellation work?
There is no fixed minimum term. Ordinary cancellation takes effect one calendar month after receipt, not only at month end. Prepaid contributions for the period after termination are returned proportionately. Mandatory legal rights remain unaffected. Full terms are provided before agreement.
Is this an investment or a charitable donation?
No equity, profit share, ownership or investment return is offered. I do not promise charitable status or tax deductibility. This is support for patent work on the terms agreed with you.
Who is responsible for the Circle?
Christina Hoffmann, acting personally, at Hofibachstrasse 3, 8909 Zwillikon, Switzerland. NOMAMIND GmbH is not the contracting party, although the page is hosted on nomamind.com.