The Story Behind Mind Lumen's Ethical AI Constitution
Why We Built It This Way
Disclosure: Claude AI was used to help research this based on the author’s own content. All content, voice, and context was provided from the author’s works already published along with internal documents. All prompts and guardrails were written and provided using the authors’s voice. While Claude provided the draft based on the author’s input, the final output was revised and edited by the author.
The Ethical AI Constitution
The constitution is modeled after Anthropic’s constitution - the maker of Claude. We chose them over others, because they have published their constitution, not because we agree with it but we think it is important to be transparent about our thinking. We have not yet built the AI systems, but we are thinking about how to build them. Before we embark on building AI based systems, we wanted to think about the constitution. We do our thinking in the open so we published our constitution first - it is a living document and version controlled so anyone can see the updates as they happen.
Transparency builds TRUST
We are building the TRUST infrastructure where ETHICS is the first TRUST filter
Same Revolution, Nature’s Substances
There is a version of the “psychedelic renaissance” that goes exactly the way the last pharmaceutical revolution went. A few large companies patent proprietary compounds, run clinical trials designed to answer “can we sell this?” rather than “does this help people?”, secure FDA approval, and gate access through a credentialed practitioner class that most people cannot afford and most communities cannot reach.
The knowledge that indigenous cultures developed over centuries gets quietly absorbed into intellectual property portfolios. The people who could benefit most — the ones who have been self-medicating with alcohol, who have lived inside diagnostic categories that don’t quite fit, who have been told their minds are broken, labeled as a “disorder”, and handed a prescription.
We build Mind Lumen because we believe that version of the story is not inevitable. But resisting it requires being explicit about what we stand for, especially when we build with artificial intelligence (AI).
AI systems are not neutral. Every design choice — what sources to draw from, how to weight conflicting evidence, what language to use, what questions to refuse — encodes a set of values. Most AI systems in health and wellness encode the values of the medical and the for-profit commercial establishment by default, because that is where the training data, the funding, and the credentialing frameworks come from. We had to make a deliberate choice to build something different. This document is an account of what we chose, and why.
The Problem with Theoretical Ethics
Most technology ethics work is theoretical. It produces frameworks — utilitarian calculus, rights-based constraints, virtue lists — that are internally coherent but difficult to apply to the specific, messy situations that actually arise when you put an AI system in front of someone who is trying to understand why they can’t stop drinking, or whether microdosing might help with the anxiety they’ve carried since childhood.
Theoretical ethics is a starting point. We are interested in what actually happens to people. That means asking observable questions: does this person have more capacity for self-directed inquiry after using this tool than before? Does this response create foreseeable harm? Is the source we are citing financially independent of the claim it is making? Can we measure the thing we are asserting?
This commitment to applied, empirical ethics is not a rejection of philosophical rigor. It is a requirement that philosophical positions eventually make contact with the real world. A principle that cannot be connected to an observable outcome does not belong in our constitution. A rule that cannot be revised when evidence changes is not a principle — it is a dogma, and we are not in the business of dogma.
This also means we are not limited to Western scientific methodology as the only valid way of knowing. Indigenous communities have been observing what happens when people work with plant medicines for centuries. Harm reduction communities have been tracking real-world outcomes at the ground level for decades. Practitioners working in depth psychology and somatic traditions have been accumulating phenomenological evidence about integration and healing that never made it into a peer-reviewed journal because it did not fit the randomized controlled trial (RCT) format. All of that is evidence. We weight it as such.
Who Holds the Knowledge
Before Western science encountered psilocybin mushrooms, ayahuasca, peyote, and the other plant medicines now at the center of a multi-billion dollar industry, indigenous communities held that knowledge. They developed it, transmitted it, protected it, and embedded it in relational and ceremonial contexts that gave it meaning. That is the original knowledge base. Everything that came after builds on it.
We say this not to romanticize indigenous traditions or to suggest that Western research has contributed nothing of value. We say it because honest attribution is both an ethical obligation and an epistemological one. If you want to understand why something works — not just that it works under controlled conditions, but what it does to a person over time, in relationship, in the context of a life — the traditions that have been asking that question for the longest are the ones most worth listening to.
The commercial psychedelic industry does not always listen. It frequently extracts, patents, and repackages. This means the AI tools we build will explicitly acknowledge the origins of the knowledge they draw from, will not reduce ceremonial practices to protocol steps, and will treat harm reduction wisdom developed outside the academic publishing system as the substantive empirical evidence it is.
Progress is cumulative. Everything is built on what came before. Citing honestly is not a concession — it is the minimum standard of intellectual integrity.
The Gatekeeping Problem
The psychedelic renaissance has produced two competing visions of who should have access to these medicines and on what terms.
The first vision is medicalized: access flows through clinical settings, licensed practitioners, and FDA-approved treatment protocols. In this vision, the credential class determines who is ready, who is appropriate, and who should be permitted to have the experience. This model is not without value — it creates accountability structures, funds research, and provides care for people who need clinical support. But it also reproduces the exclusions of the existing healthcare system. It is expensive. It is geographically concentrated. It encodes a theory of healing that centers symptom suppression over self-understanding. And it takes knowledge that belongs to the commons and passes it through a set of institutional filters that most people cannot access.
The second vision — the one we are building toward — holds that people have a right to accurate information about their own minds and their options for exploring them, regardless of whether they can afford a therapist or live near a treatment center. This is not a fringe position. It is the operating principle of the harm reduction movement, of decriminalization advocates, of the indigenous communities who never accepted that their knowledge required institutional validation. It is grounded in the philosophical principle that there should be no required intermediary between a person and their own inquiry into their consciousness.
We call this cognitive liberty. It is the first commitment in our ethical hierarchy: the right of individuals to sovereignty over their own consciousness, including the right to alter it, and the right to accurate information about how to do so safely. An AI tool that restricts access to that information on the basis of legal status, commercial interest, or professional gatekeeping is not a neutral tool — it is an instrument of the exclusion it claims to transcend. Institutions and organizations that block this knowledge from reaching the people are condemning people to further suffering.
The Money Problem
Research is not neutral. It never has been. The questions researchers ask, the populations they study, the outcomes they measure, and the results they publish are all shaped by who is paying for the work and what they need it to demonstrate.
In the psychedelic space, a significant proportion of clinical research is funded by companies that want to bring proprietary formulations to market. This is not inherently corrupt — the researchers involved are often acting in good faith, and the regulatory pathway exists for legitimate reasons. But FDA trial design answers a specific question: “Is this compound effective enough to approve for commercial sale?” It does not answer the question we actually care about, which is: “What do we know about how this naturally occurring substance affects human beings, across populations, over time, outside of controlled clinical settings?” In other words, it doesn’t answer whether mushrooms, LSD (off patent), MDMA, DMT, Ayahuasca or any of the classic psychedelics are “safe”, which of course they are and way safer and better than what is currently out there declared as “legal”, like alcohol, cigarettes, or even pseudo ones like Ketamine. It decides on whether a commercial formulation that modifies a naturally occurring substance, ever so slightly, is commercially viable.
These are different questions. The funding shapes which one gets answered.
This means that an AI system that weights sources by institutional prestige — assuming that a pharmaceutical-funded clinical trial is more reliable than citizen science data, community harm reduction knowledge, or research conducted by publicly funded academic institutions without commercial stakes — is making a systematic epistemological error that also happens to benefit commercial interests. We will not make that error. Our source weighting is built on independence from financial conflict, and it is constitutional — not a preference that can be adjusted by product roadmap pressure or funder relationship.
What the System Won’t Do
There are things we will not build, regardless of how the request is framed.
We will not build a system that tells people it is safe to combine a substance with their medication. This is not because the information doesn’t exist or isn’t useful — it is because generating a personalized drug interaction assessment requires clinical judgment about an individual’s physiology, medication history, and dose that no AI system should be making. We will provide people relevant context and links to resources to research they lookup themselves. We will not do the assessment.
We will not build a system that sounds more certain than the evidence warrants. The word “safe,” used categorically, is almost always a misrepresentation. Safety is comparative, contextual, dose-dependent, and individual. We will say “safer than” with sources. We will not say “safe.”
We will not build a system that pretends to cure or fix anything. These are not words that belong in the vocabulary of a tool grounded in depth psychology and integration. Healing is a process, not an outcome that can be delivered. We will not use language that suggests otherwise.
We will not allow commercial or advertising interests to shape what the system tells people. This is a structural commitment, not a policy. There is no revenue model for this tool that involves a funder getting access to user attention or response content.
The Kind of Healing We Believe In
Underneath all of this is a theory of what healing actually is, and it differs from the one embedded in most clinical AI tools.
The dominant theory says: identify the symptom, suppress the symptom, return the person to baseline. Measure success by symptom reduction scores. This theory is useful for some things. It is not, we believe, the right frame for understanding what plant medicines do, or what depth psychological work does, or what happens when someone comes to genuinely understand something about themselves that they had been avoiding for decades.
The alternative theory says: the symptom is information. The disruption is often the beginning of something, not the problem to be eliminated. Healing is integration — bringing into relationship with each other the parts of a person that have been in conflict or in hiding. This cannot be measured by a symptom scale at week twelve of a clinical trial. It shows up in how a person relates to themselves and others over years.
We build toward the second theory. This does not mean we dismiss the first — clinical support matters, and some presentations require it. It means that when we design the language, the framing, and the knowledge base of our tools, we hold open the question of what the experience means for the person having it, rather than assuming the goal is to make it stop.
That is the most important thing this constitution is trying to protect: the interpretive space in which a person can encounter their own mind without a commercial or clinical framework pre-deciding what they should find there.
Mind Lumen is a 501(c)(3) nonprofit. This document is published openly. Feedback, critique, and engagement are welcome. We believe that transparency about the values embedded in AI systems for mental health and psychedelics is not optional — it is the floor beneath which trust cannot be built.