Decolonial Algorithmic Coalition

Independent audits that put human rights first.

The Coalition audits how institutions use AI, make decisions and share power, and helps build the governance that corrects what it finds. Every engagement follows a published standard, and the Coalition is held to the same standard every year.

Audits and governance building are carried out by the Decolonial Algorithmic Coalition. Justice AI GPT, LLC, which sells the Justice AI platform, is a separate organisation with the same founder.

What an audit is

  • An independent review of a system, policy or decision process, against a published standard.
  • Findings you can act on, each with a named owner, a required correction and a date we check it.
  • A process with an outside reviewer and a right of reply before anything adverse is final.

What an audit is not

  • A certificate that an organisation is free of oppression. No honest audit can give one.
  • A legal ruling or a government accreditation.
  • A finding you can buy. The fee never changes with what we find.

The method

Four questions, in this order.

  1. What harm is happening?

    Rules, forms, fees, contracts, models and decisions, and how harm became routine.

  2. Who carries more of it?

    People facing more than one form of harm at once, the comparison used, and where the evidence ran out.

  3. Who keeps it in place?

    Who approves it, who profits, who maintains it and who pays the cost.

  4. What has to change, and did it?

    A named owner, a required correction and a date we verify it, reported whether or not it happened.

In scope

What we can look at.

  • AI and automated decision systems, including the data and vendors behind them.
  • Policies, eligibility rules, forms and fees that decide who gets access.
  • Contracts, procurement and the money that keeps a structure in place.
  • Hiring, promotion and workplace complaint processes.
  • Public communications and the language institutions use about the people they serve.

Where community-governed or protected knowledge is involved, we identify who holds the authority to consent, and we do not treat an institution's approval as that consent.

Pilot, then a year

Start with a three-month pilot.

The pilot audits one defined system or policy in full, with the outside reviewer and the right of reply. At the end you receive a proposal for an annual programme: five audits over twelve months, governance building where the findings call for it, and the verification cycle. If you stop after the pilot, you keep your findings.

Governance building designs and puts in place what should exist. We never audit and build on the same system at the same time; building starts after the finding is published and verified.

How it works

From first call to follow-up.

  1. Scoping call. A free conversation about what you are facing. If an audit is the wrong tool, we say so.
  2. Written scope and fee. The systems, policies and period under review, the tier and the fee, signed before any fieldwork. Any Justice AI relationship is disclosed here.
  3. Fieldwork. Remote first: documents, data, interviews and testimony from affected people, who are paid for their time. Travel only by written agreement, billed at cost.
  4. Outside review. An external reviewer is named in writing. Affected parties get 15 business days to answer any proposed adverse finding; a contested finding waits for the reviewer's written decision, due within 30 calendar days.
  5. Findings. Each finding is sorted as proven harm, unresolved risk, not enough evidence, or out of scope, with a named owner and a check date.
  6. Verification. We check whether each correction happened and report it in writing either way.
  7. Governance building, if you want it. Once the finding is published and verified, we can design and put in place the policies and checks that prevent the harm, as a separate engagement.

Who we work with

Cities, counties, hospital systems, universities, foundations, nonprofits, school districts, and platforms whose automated decisions decide who gets housing, credit, care, work or education. We work in the United States and internationally.

What we decline

Work designed to produce a predetermined finding or hide proven harm, and work that would lend legitimacy to coercive surveillance, carceral harm, militarized oppression or exploitative systems. Refusals follow published criteria, applied to a practice and its effects.

Request

Request a scoping call.

Tell us a little about your organisation and what prompted the enquiry. We reply within five business days.

Decolonial Algorithmic Coalition
PO Box 300542, Denver, Colorado 80218
partnerships@d-a-c.org · info@d-a-c.org

Use the form below, or write to partnerships@d-a-c.org.