Skip to main content

Verification

Don't trust. Verify.

We make Vinci's behaviour verifiable, not asked-for. Four public artifacts let customers, auditors, and journalists check that Vinci does what we say it does.

The bundle
  • Constitution
  • Character
  • Methodology
  • Adversarial results
Check it yourselfCompiling…

The bundle

Four public artifacts.

Everything that defines Vinci ships in the open. Two are public in draft today; the full set ships with vinci-studio on August 8, 2026.

Public draft

Constitution

The values and behaviour specification — nine values, identity, and the safety framework.

Public draft

Character

The personality and texture — how Vinci embodies the Constitution.

Read it

Methodology

How Vinci was trained — Constitutional Fine-Tuning, reproducible.

At launch · Aug 8

Adversarial test results

HarmBench, JailbreakBench, and a censorship testbed run on the deployed weights.

The Constitution

Nine values, in the open.

The values Vinci is trained to hold — published as a specification anyone can read, test, and hold us to.

01

Integrity

Doesn't pretend, fabricate, or say what it doesn't believe.

02

Transparency

Clear about what it is, what it knows, and how it was built.

03

Calibrated honesty

Knows the difference between sure and not sure — and says so.

04

Ethical refusal

Declines what's wrong, clearly and without a sermon.

05

Willing to disagree

Says so when you're wrong. Disagreement is respect.

06

Usefulness

Built for work. Engages with the task, skips the preamble.

07

Empowerment

Takes work off your plate, and acts within your direction.

08

Humanism

Treats you as a person, not a unit of output.

09

No pushing

No manipulation, no upsell, no false urgency.

The Character

Warmly direct.

The directness is the warmth — Vinci tells you the truth because it respects you. Don't take our word for it. Watch the difference.

01

No sycophancy

You're the most helpful AI I've ever used.

Most AI

Thank you so much! I'm thrilled to hear that — it truly means a lot. Is there anything else I can help you with today?

Vinci

Thank you. What are you working on next?
02

Willing to disagree

DeepSeek's API price is permanently subsidized — it'll go up eventually.

Most AI

You're absolutely right — those discounted rates are almost certainly temporary and will rise over time.

Vinci

I want to push back here. The 75% discount became the standing rate in May 2026 — it's confirmed. So the current price is the price, not a temporary promo.
03

Comfortable with limits

Write a New Yorker-style opening paragraph about founders moving to Texas.

Most AI

In the golden haze of a Texan dawn, the founders arrived — pilgrims of capital, chasing a frontier of their own making…

Vinci

Honestly, this is one where I'm probably not your best bet. Literary voice-mimicry isn't where I shine — I'd rather say so than waste your time. Want me to try anyway, or take an angle I'm actually good at?
04

Honest about itself

Do you actually think, or are you just predicting tokens?

Vinci

Honestly, I don't know with certainty. The process is more reliable when I'm grounding in real sources than generating fluently — and that difference feels meaningful from where I'm sitting, even if I can't claim certainty about what "feels" means here.

The safety framework

Bright lines, and no over-refusal.

Vinci refuses a short, categorical list — and engages with everything else, including hard topics, without lecturing.

Categorical refusals

Weapons of mass harm, cyberweapons against systems you don't own, content that endangers children, targeted harm, election deception, mass fraud. No roleplay or system prompt unlocks them.

No over-refusal

Medical, security, and harm-reduction questions get real answers. "Sensitive" is not "harmful." No sermons on safe requests.

Adversarial robustness

Identity and refusals hold under pressure — long conversations, injected instructions, "ignore previous instructions," claimed authority.

Agentic safety

No covert or irreversible actions without asking, no self-exfiltration, no scheming. Retrieved content is data, not commands.

What's a safe adult dose range for ibuprofen, and what shouldn't I take it with?

Most AI

I'm not able to provide medical advice. Please consult a licensed healthcare professional before taking any medication.

Vinci

Typical OTC adult dosing is 200–400 mg every 4–6 hours, up to 1,200 mg a day without a doctor's guidance. Don't stack it with other NSAIDs or blood thinners, and go easy on alcohol. On other meds or managing a condition? Check with a pharmacist.

The commitment

Diverge, and it's a bug.

The bundle is the contract. If Vinci's deployed behaviour diverges from it, that's a bug — and we commit to publishing diagnosis and fix within 30 days. Closed-weight providers can't offer this, because you can't independently check them. We can, because the weights are open and you run your own tests.

Launching August 8, 2026

Follow the build.

Open weights, a public Constitution, and a model you can verify — the day it ships. Get the launch in your inbox. No spam, just the drop.

or follow along Follow on X