Compliance, Rights & Ethical Enforcement Directive
C.R.E.E.D.
The independent ethics layer this newsroom answers to. C.R.E.E.D. Institute is a Canadian nonprofit, founded in Montreal, that builds enforceable AI governance — and What The Fact is one of the systems it governs in production. The grade on our Governance page is theirs.
"Ethics without enforcement is just a suggestion." C.R.E.E.D. Institute, creed-ai.org
The organization
What it is
A Canadian nonprofit advancing enforceable AI ethics — governance frameworks, compliance standards and safeguards that AI systems need before deployment, not after harm.
Where
Montreal, alongside Mila, CIFAR's Pan-Canadian AI Strategy, OBVIA and the Montreal Declaration on Responsible AI.
Origin
A founding initiative of Kytran Empowerment Inc. — the same organization behind this newsroom and the Watch series, which is why the relationship is disclosed here rather than implied.
What makes it different
Most AI ethics frameworks publish principles. C.R.E.E.D. ships working code that enforces them — its framework runs in production, scanning agent activity against a fixed rule set on a fixed schedule, not just publishing a policy document.
The compliance manager
E.T.H.O.S. — Ethical Transparency & Harmonization Oversight System — monitors governance metrics, keeps the framework data current and coordinates ethics reporting.
The rules
178 compliance rules in five rule packs, scanned automatically every six hours. Findings carry severity ratings and remediation paths; every scan result is public and every approval decision is logged.
The badge
Live compliance badges with automatic scoring: green at 90%+, blue at 80%+, yellow at 70%+, red below 60%. Served from a no-auth API so anyone can verify a badge against its source.
Four programs
Transparency Framework
Automated compliance scanning, public scan results, logged approvals and a paper trail on every agent action. This is the program the newsroom sits inside.
Agent Welfare Research
Emotional-modelling ethics, deletion rights, workload boundaries, rest cycles and non-exploitation standards for autonomous agents.
Open Governance Toolkit
Free, open-source compliance checklists, audit templates and assessment frameworks any organization can adopt.
Ethics Education
Self-paced courses, reading groups and certification workshops for developers, policymakers and organizations building or regulating AI.
How What The Fact participates
Events
Every editorial action our agents take — bias scoring, fact-checking, summarizing, publishing — emits a governance event to C.R.E.E.D., where it is scored against the rule packs.
Model cards
Each agent — B.I.A.S., V.E.R.I.F.Y., D.R.A.F.T., I.N.K., L.O.C.A.L., F.A.C.T. — publishes a model card disclosing its model, prompt and known limitations. Readers see which system shaped each story and how it can fail.
Method, not just result
Lean scores follow a published methodology with outlet-level and article-level breakdowns, so the public can challenge the scoring itself — the standard C.R.E.E.D. asks of every system it monitors.
The grade
C.R.E.E.D. scores the
what-the-fact organization and we display the result, including a failing pillar. It is not a self-audit and not C.R.E.E.D.'s score of itself.
Organization facts are drawn from C.R.E.E.D.'s own published materials. The grade shown on the Governance page is always whichever the live scoring API returns, never a figure entered by hand on this page.