Evidence before confidence.

How the legacy score works

Score = bias mitigation × 25% + regulatory alignment × 25% + transparency × 20% + explainability × 15% + safety × 15%, rounded to one decimal. Inputs are inherited editorial estimates. Historical evaluation records and review dates are unavailable; these are not reproducible benchmarks.

For continuity, scores ≥85 are labeled “Higher editorial estimate”; 60–84.9 “Mid-range editorial estimate”; below 60 “Lower editorial estimate”. These labels do not establish certification or compliance. Default sorting is alphabetical.

What evidence labels mean

Documented support

A dated, product-specific source describes support for a requirement. This does not establish legal conformity.

Partial support

A source documents some capabilities; gaps are stated.

Not verified

No sufficient source has been checked and recorded. Every current framework mapping has this status.

Not applicable

A documented scope assessment explains why the requirement does not apply.

What a verification record needs

Product and plan, exact claim, source URL, supporting excerpt, scope and exclusions, reviewer and check date. Certification claims also require the issuing body, certificate identifier, validity and certified organizational scope. ISO/IEC 42001 concerns an organization's AI management system; a software feature badge is not a certificate.

Prices and review dates

Unsubstantiated inherited prices have been withdrawn. Platform prices and framework evidence are not verified. The interface and methodology were revised on 3 November 2026; this is not a vendor evidence review date.

IBM product distinction

The former $600/month OpenScale claim has been removed. IBM lists watsonx.governance Risk & Compliance Basic from $3,500/month, a different product package. Do not treat this as an OpenScale quote. IBM pricing → (checked 3 November 2026).

EU AI Act timeline

Annex III high-risk requirements apply from 2 December 2027; relevant Annex I product requirements apply from 2 August 2028. Other obligations apply earlier. See the European Commission's current timeline (checked 3 November 2026).

Human oversight requires people and procedures

Article 14 includes understanding system limits, recognizing automation bias, interpreting outputs, and the ability to disregard, override or stop a system where appropriate. Monitoring software alone does not satisfy these requirements. Read the AI Act →

Independence and scope

This is a discovery directory, not legal advice or a certification service. No new hands-on evaluations were performed for this redesign. Provider links are starting points for research, not supporting evidence for every feature. Existing site sponsorships do not count as evidence.