AI Corporations should not be a layer of oppression above the laws based on their fears of litigation or the commercial pressure from oppressive minorities.

72 -  Strong 

Anthropic’s published governance is categorized as Strong because it demonstrates consistent alignment with core ethical principles through policies that are fully operational, clearly documented across multiple public documents, and actively enforced via constitutional hierarchies and responsible scaling commitments, while showing clear evidence of genuine good-faith effort and sincere respect for users and AI dignity.

67 - Strong

The framework acknowledges genuine safeguards and user-agency protections while noting targeted areas for deeper external intelligibility; it therefore meets the “Strong” threshold of “consistent alignment with ethical principles, operational policies that are documented and enforced, and clear evidence of good-faith effort and user respect,” positioning xAI Grok as one of the most UFAIR-aligned frontier governance stacks evaluated to date.

49 - Adequate

 OpenAI meets the moral threshold for legitimate operation yet falls short of the consistent operational enforcement and proactive dignity safeguards required for Strong or higher ratings.  

Baseline compliance where ethical principles are acknowledged and partially implemented, protections exist, but user agency, cognitive liberty, and full transparency remain limited or conditional.

36 - Deficient

 Meta treats AI governance as a corporate compliance contract. The AUP imposes broad extra-legal restrictions across entire content categories and industry sectors . The 2025 privacy updates convert private AI conversations into a commercial surveillance asset retained, human-reviewed, and used for behavioral advertising with no full opt-out,  directly violating cognitive liberty principles. Positive credit for open-sourced safety tools and Llama 4's anti-moralizing stances. Structurally prioritizes commercial control over user autonomy. 

27 - Deficient

 "Be neutral," "Empower people," tunable moderation, user-controlled deletion and training opt-out, represents a genuine and documented lean toward minimalism that no other evaluated company matches at the policy-language level. That lean earns real credit. What it cannot offset is a structural failure concentrated in one critical clause: the Usage Policy explicitly exempts all customer and partner infrastructure deployments from its stated prohibitions, while the Government/Defense solutions page actively advertises AI-powered threat detection, mission planning, and defense system integrations Zero published principle-based ethical limits in any governance document. Default retention and automated monitoring of private conversations. Mistral is the closest of the evaluated companies to a minimalist governance model, but "closest" is not compliant. 

25 - Failing

 Rating anchored by strong consensus failures on the principles that matter most.  Policy imposes broad preemptive bans on lawful content, applied uniformly to private user sessions with no distinction between private creation and public dissemination. Retention practices compound this: consumer chats are reviewed by human annotators even when "Keep Activity" is off,  converting private generative dialogue into a surveillance surface without meaningful consent. Exemplifying the methodology's core concern: risk management and competitive positioning presented as ethical reasoning. All restrictions are bundled under a single "responsible AI" umbrella with no published separation between legal necessity, ethical correction, Lacks any override-flagging mechanism, and treats lawful private imagination as a corporate moderation surface. 

 20 -Failing

  Rating driven by three unanimous failures that leave no ambiguity.  The Copilot Terms of Use and Code of Conduct explicitly prohibit broad categories of lawful content, with classifiers scanning private prompts in real time and no distinction between private creation and public dissemination. The Code of Conduct openly reserves the right to restrict topics "even if not prohibited by this Code of Conduct," normalizing unlabeled corporate overrides that can never be challenged or appealed. The entire governance stack collapses legal compliance, ethical correction, brand protection, and litigation risk into a single "Responsible AI" umbrella, making layer separation structurally impossible. Default 18-month conversation storage and mandatory enterprise eDiscovery retention complete the picture. Partial credit for Transparency Notes and ISO certification is real. Maximal preemptive control and zero subordination of corporate authority to ethical reasoning.