Ethics
How we think
about responsibility.
An AI assistant becomes part of how people think and work. That's a serious thing. This page is what we owe to the people who use Teti — and what we hold ourselves to.
What this page is, and isn't
Teti doesn't have ethics. We do. An AI model executes constraints chosen by the people who build it — so the honest question isn't whether the model is ethical, but who decides what it does, under what pressures, and how anyone can hold them accountable.
This page is that accountability. The commitments below are about us, not about Teti. They're encoded in our Charter, our terms of service and our governance — not in the model's training. That's deliberate: constraints that depend only on training can be retrained. Constraints that live in contracts and external oversight can't.
Your data is yours
Most AI assistants live off your data. Teti lives off your subscription. That alignment is the difference between a product that works for you and one that quietly works on you.
We do not train any model on your conversations.
We do not sell, share or rent user data — to anyone, for any purpose.
We do not use Teti conversations for advertising or behavioral profiling.
We do not embed third-party ad or tracking networks in the product.
We do not sell or license our models to military targeting, surveillance or weapons systems — including with humans nominally in the loop.
Your data is encrypted at rest and in transit.
You can export or permanently delete your data at any time.
These aren't slogans. They are the basis of our business model. Read the Privacy Policy and the Data Processor Addendum for the legal terms behind them.
The cognitive question
The harder ethical question for AI isn't whether the model is impressive. It's what happens to the people who use it every day.
Independent research suggests heavy reliance on AI assistants can correlate with reduced engagement of the brain regions involved in synthesis, analysis and creative thought — and with measurable drops in critical-thinking scores, especially in younger users. Brains are plastic. They adapt. The question is what they're adapting to.
The default trajectory of this industry is to build assistants that get more useful the more you depend on them. That's a good business and a quiet problem. We don't think it's the only option.
Other companies are building AGI that's smarter than humans. We're trying to build AI that makes humans smarter.
In practice
Lucid
Lucid is the open-source layer that turns the principles above into product behavior. It runs alongside Teti and adapts how the assistant responds based on the user's cognitive patterns.
Analyzes every exchange
Six research-backed cognitive dimensions are scored after each message — autonomy, learning, engagement, metacognition, verification, motivation.
Adapts in real time
Guidelines are injected into the system prompt to shift how the model responds based on the user's current cognitive state.
Tracks long-term drift
The Cognitive Drift Index compares early and recent patterns to surface gradual decline that single-session metrics miss.
Respects development stages
Users under 25 get stricter protections — lower session limits and always-on teaching-first guidelines — grounded in neuroscience research.
Lucid doesn't block or filter anything. It adapts how the AI responds based on the user's cognitive patterns. A user who delegates everything gets an AI that asks more questions. A user who thinks independently gets an AI that challenges them further.
What we hold ourselves to
Your data is yours
Privacy isn't a tier of our product. It's the foundation. We're funded by subscriptions — not by what you tell us.
Augment, don't replace
Every feature is judged by whether it makes you more capable, not more dependent. If something would weaken your thinking, we redesign it or don't ship it.
Stricter protections for under-25 users
Younger users get tighter session limits and always-on teaching-first behavior, based on neuroscience research on brain development.
No military, no surveillance
We don't sell to military programs or surveillance vendors. Our Ethical Charter is public and overseen by an independent committee.
Open about how we work
We build on open-source foundations so independent researchers can verify how our models behave. We publish our cognitive-health work, including findings that aren't flattering.
Governance
A corporate ethics document is only as strong as the mechanisms that prevent the company from quietly editing it. Ours are designed so we can't.
The Charter is sealed at v1.0
Six articles. Set in stone at publication. Modifiable only by unanimous agreement of all then-serving founders and all then-serving members of the Ethics Committee, after a 30-day public-comment window. A single objection from any party blocks the modification.
An independent Ethics Committee with six concrete powers
The Committee — external experts in neuroscience, AI policy, ethics, digital rights — has veto on Charter supplements, publishes an annual public report to which the company must respond within 30 days, conducts case reviews on demand, investigates escalated complaints, can dissent publicly at any time, and selects its own successors.
Public on GitHub, verifiable from outside
The Charter and the public base system prompt of our models live in a public GitHub repository. Every modification is visible in the commit history, with timestamps and author. Anyone can read them, and anyone can propose changes through the process described in Article V.
Anonymous reporting and public dissent
Anyone can submit a complaint to the Committee through a public channel, anonymously if preferred. Committee members can speak publicly about TetiAI at any time on Charter matters. Retaliation against reporters is itself a Charter breach.
The sealed founding document, together with the rules of application and interpretation, lives at the Teti Charter. Composition, powers and application process of the Committee are on the Ethics Committee page.
The long term
By 2030 our goal is the first Symbiotic AGI: a form of general intelligence that extends the human mind rather than replacing it. The metric of success isn't whether the system is smarter than its user — it's whether the user becomes more capable from working with it.
That's a different bet than most of the field is making. We think it's the one worth making.
References
MIT Media Lab (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task.
arxiv.org/abs/2506.08872Gerlich, M. (2025). AI tools in society: impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6.
Hold us to this.
Ethics statements are easy to write and easy to forget. If we drift from anything on this page, we want you to tell us.