Responsible AI.
The limits we put on ourselves. Stated here so they can be held against us.
Draft — not yet live. This page states the commitments the rest of the site already makes. The operative details are marked below and it needs review by counsel before launch.
What we will not build
We do not build systems whose purpose is to manipulate attention or behaviour against a person’s interest. No engagement mechanics, no artificial urgency, no design whose success is measured by how long it kept someone rather than what it did for them.
Training and consent
We do not train on customer data without explicit, revocable permission. Revocable means it can be withdrawn and the withdrawal means something — which is a data lineage problem, and we treat it as one.
Being wrong
A system that is confidently wrong is a defect, not a limitation. When the system is unsure it says so, and calibrating that is engineering work with an owner, not a disclaimer.
Any decision with a material consequence carries a trail of what informed it, designed in rather than retrofitted.
Uneven failure
Systems fail unevenly, and the people they fail hardest are usually the least able to argue with them. We test for that specifically, and we treat a disparity in failure rate as a bug with a severity, not a footnote in a model card.
Telling you when we get it wrong
If a system of ours causes material harm, we will say so, say what we changed, and say it in public rather than in a support ticket.
Questions about this page.
Write to hello@the01x.com. Security researchers should use security@the01x.com — we answer those first.