We gave it a conscience
before we gave it a face.
Artificial intelligence is powerful enough to reshape society. That power demands a framework — not a disclaimer in the footer of a terms-of-service page. This is ours.
Narrow AGI.
Scoped intent.
No unbounded autonomy.
We don't build toward general intelligence. We build toward narrow, purposeful systems that serve a defined function and nothing more. Every capability we ship is scoped to a specific intent — and that scope is enforced, not aspirational.
Autonomy is a tool. We use it deliberately, and we never let it run ahead of accountability.
not general
scoped intent
values in
human oversight
Five promises. No asterisks.
No deepfakes of real people without verified consent.
We will not synthesize a likeness of a real person — voice, face, or mannerisms — without a clear, affirmative consent process. Not from a public figure. Not from a celebrity. Not from anyone.
No emotional manipulation patterns.
No sycophancy loops. No dependency triggers. No designed parasocial attachment. We build AI that informs, not AI that grooms. If a pattern increases engagement at the cost of the user's agency, we remove it.
Transparent AI disclosure on every interaction.
Every user knows when they're speaking with a digital human. We don't obscure the AI nature of our interactions. We don't design for deception. Disclosure is a feature, not a compliance burden.
User-owned voice and likeness data. Exportable and deletable.
Your voice. Your face. Your data. We don't hold rights to what you create with our platform. Export your data anytime. Delete it anytime. We don't sell it, license it, or train on it without explicit opt-in.
Public incident log when we get it wrong.
When something breaks badly — a misuse case, a failure of alignment, a harm caused by our system — we log it publicly. Not a press release. Not a legal statement. A real incident log with what happened and what we did about it.
What we will not build.
These aren't edge cases to debate. They're hard limits.
Impersonation for fraud
Synthesizing a real person's likeness to deceive financial institutions, bypass security, or impersonate someone in a transaction.
Persuasion for political operations
AI-generated persuasive content designed to influence elections, public policy, or civic outcomes — at scale, without disclosure.
Parasocial-by-design
Systems explicitly engineered to simulate human intimacy, friendship, or romantic attachment to extract engagement or revenue. Connection that isn't real is a product defect.
This document is reviewed annually and updated when our thinking evolves. The commitments above reflect our current operating principles — not aspirational marketing copy.