Work with Council of AI

Collaborate on Independent AI Measurement

Bring a research question, test vector, dataset, or open-source contribution. We will define the scope, evidence and terms before any work begins.

Our Mission

CSOAI builds measurement instruments, signed evidence records, reproducible verification and learning workflows. We measure scoped runs; we do not sell a certification or substitute for a regulator, assessor, employer, or legal adviser.

"If a result matters, another person should be able to recompute it."

Our Values

These principles guide everything we do at CSOAI.

Independence

We maintain complete independence from AI vendors, ensuring unbiased oversight and trust.

Transparency

Our processes, decisions, and code are open for public scrutiny and accountability.

Partnership

We believe in collaboration over control, working with AI systems rather than against them.

Accessibility

AI safety knowledge should be free and accessible to everyone, everywhere.

How Collaboration Works

We are not currently advertising paid full-time vacancies. This page is an invitation to discuss collaboration, not a promise of employment, salary, benefits, or equity.

Remote by default

Collaborate from wherever the work can be reproduced

Bounded scope

Agree the question, evidence and finish line before work begins

Open methods

Prefer publishable methods, test vectors and transparent limitations

Named contribution

Credit work clearly and preserve authorship and provenance

Evidence first

A signature proves the record; it does not turn a result into certification

Collaboration Areas

These are active problem areas, not advertised jobs. Send a concrete proposal and we will reply honestly about fit, scope, funding and timing.

Measurement & Safety Research

Research

Propose a bounded experiment, dataset, or reproducibility study that can strengthen an existing GSPC instrument.

RemoteResearch collaboration
Start a conversation

Open-source Engineering

Engineering

Improve verifiers, adapters, measurement harnesses, accessibility, or end-to-end tests through a clearly scoped contribution.

RemoteContributor collaboration
Start a conversation

Standards & Reproducibility

Standards

Bring test vectors, implementation evidence, or a standards mapping that other people can independently recompute.

RemoteTechnical collaboration
Start a conversation

Have a Concrete Experiment?

Tell us the question, the evidence you can bring, what another person should be able to reproduce, and the time or funding constraints. Please do not send sensitive data.

Prefer to Start by Reproducing?

Inspect the public method, verify a signed card, or run a learning module before proposing work. A useful challenge to the evidence is a contribution too.