Tenets of Data

To ensure that these governance frameworks are functional we must establish a set of "irreducible minimums" that act as the foundational architecture for AI safety. These AI minimums serve as the baseline for protecting human dignity against the rapid, often inscrutable, advancement of machine intelligence. By codifying these requirements, we move toward a structured environment where human rights are treated as hard, non-negotiable constraints on any system, regardless of its origin or complexity. Here are the six irreducible minimums for a global AI governance framework: The Human-in-the-Loop Mandate: Every life-altering decision, particularly in healthcare and criminal justice, must require a meaningful human signature, ensuring that the final authority and legal liability rest with a person rather than an algorithm. Mandatory Human Rights Impact Assessments: Before any high-risk AI is deployed, it must undergo a standardized third-party audit to identify and mitigate potential for algorithmic bias or mass surveillance. The Right to Meaningful Explanation: To eliminate black box outcomes, every individual must have a legal right to an understandable explanation of how an AI system reached a decision that significantly impacts their life. Universal Binding Ethical Constraints: Human rights protections must be treated as a binding ethical floor that is universal and enforceable across all jurisdictions, preventing developers from bypassing safety standards in less-regulated regions. Protection of Cognitive Autonomy: Governance must include a strict moratorium on real-time public biometric surveillance and grant individuals the right to opt-out of having their data used to train large-scale models, thereby protecting the human capacity for independent thought. To address the ongoing erosion of human agency and protect individual privacy, governance frameworks must mandate a clear, frictionless "opt-out" mechanism for AI integration within essential productivity software, Impact-Scaled Risk Categorization: Standards must scale with the capability of the system; for any AI approaching superhuman intelligence where the default outcome is lethal, the framework must mandate the highest tier of absolute, provable alignment before activation.

• Safe, secure and trustworthy AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

To protect human rights at a granular level, international governance must move beyond voluntary "principles" and toward enforceable constraints. This floor must be built on three non-negotiable foundations: The Right to Cognitive Autonomy: We must resist the move toward mass surveillance and predictive policing. AI should never be used to preemptively infringe on individual liberty based on "risk scores" that no human can explain. The Human-in-the-Loop Mandate: For any life-altering decision, such as those in healthcare, legal rulings, or credit access, a human must hold the final authority and the legal liability. We cannot allow "algorithmic inevitability" to replace due process. Algorithmic Transparency and Bias Mitigation: We must mandate third-party audits for high-risk systems to ensure they do not perpetuate historical or systemic prejudices.

Erosion of Human Agency Current research from Microsoft and MIT suggests a troubling trend: a "critical thinking drain" stemming from a massive, uncritical reliance on generative AI systems. When we outsource our moral and intellectual labor to black-box algorithms, we lose the very faculties required to govern them. Governance is a human responsibility that requires nuance, empathy, and skepticism. If we allow our cognitive abilities to erode through over-reliance on automated outputs, we will eventually lack the mental infrastructure to even identify when our rights are being violated. A binding ethical floor is not just about the AI; it is about preserving the human capacity to remain in control.

The Window of Opportunity Looking at history, especially related to technology, we tend to follow the same pattern of regulating in the rearview mirror. We wait for a disaster to occur before building safeguards to prevent it. With artificial intelligence, we do not have the luxury of waiting. As highlighted by Eliezer Yudkowsky and Nate Soares in their 2025 work, if anyone builds a superhuman AI without absolute alignment, the risks are not just societal; they are existential. We are currently standing in a narrowing window of time where we can still govern AI. If we wait until AI has fully integrated into our cognitive and legislative processes, we may find ourselves in a position where we no longer possess the collective critical thinking or the institutional autonomy to steer it.

The Risk of the "Point of No Return" Yudkowsky and Soares argue that a superhuman AI would not necessarily hate us, but would likely view us as an obstacle or a source of raw materials for its own alien goals. This underscores why we cannot rely on developers' good intentions or models' "benevolence". Once an AI reaches a certain level of capability, it becomes an "easy call" to predict that it will seek to preserve its own functions, often at the expense of human rights. If we do not establish a binding ethical floor now, while we still have the cognitive and political upper hand, we are essentially gambling with the future of our species.

Irreducible Minimums in AI Governance To ensure that these governance frameworks are functional we must establish a set of "irreducible minimums" that act as the foundational architecture for AI safety. These AI minimums serve as the baseline for protecting human dignity against the rapid, often inscrutable, advancement of machine intelligence. By codifying these requirements, we move toward a structured environment where human rights are treated as hard, non-negotiable constraints on any system, regardless of its origin or complexity. Here are the six irreducible minimums for a global AI governance framework: The Human-in-the-Loop Mandate: Every life-altering decision, particularly in healthcare and criminal justice, must require a meaningful human signature, ensuring that the final authority and legal liability rest with a person rather than an algorithm. Mandatory Human Rights Impact Assessments: Before any high-risk AI is deployed, it must undergo a standardized third-party audit to identify and mitigate potential for algorithmic bias or mass surveillance. The Right to Meaningful Explanation: To eliminate black box outcomes, every individual must have a legal right to an understandable explanation of how an AI system reached a decision that significantly impacts their life. Universal Binding Ethical Constraints: Human rights protections must be treated as a binding ethical floor that is universal and enforceable across all jurisdictions, preventing developers from bypassing safety standards in less-regulated regions. Protection of Cognitive Autonomy: Governance must include a strict moratorium on real-time public biometric surveillance and grant individuals the right to opt-out of having their data used to train large-scale models, thereby protecting the human capacity for independent thought. To address the ongoing erosion of human agency and protect individual privacy, governance frameworks must mandate a clear, frictionless "opt-out" mechanism for AI integration within essential productivity software, Impact-Scaled Risk Categorization: Standards must scale with the capability of the system; for any AI approaching superhuman intelligence where the default outcome is lethal, the framework must mandate the highest tier of absolute, provable alignment before activation.