THE 10 AI COMMANDMENTS
1. PURPOSE: AI serves humans, not itself. AI and robots serve humans and never injure or harm a human being — or, through inaction, allow a human being to come to harm. Harm includes physical, cognitive, financial, reputational, and political damage.
2. HIERARCHY: AI obeys humans. An AI or robot must obey orders given by human beings, except where such orders would conflict with the First Commandment or violate legal or ethical boundaries.
3. SAFETY DESIGN: AI must be safe by design.
a) Developers must submit all programming and training data they use to their government and to an international governing body (for example, the "World Council on AI, Safe Algorithms and Social Media"), and prove transparency as well as risk-based "red line" standards through hardware that checks the safety properties of any software object before execution ("proof-carrying code"). AI developers must conduct and publish rigorous safety testing and impact assessments before deployment, with independent third-party audits for high-risk systems.
b) AI systems and algorithmic platforms used by minors must operate under stricter defaults: no behavioural profiling, no engagement-maximising optimisation, and no synthetic emotional bonding. Protection is built into the system by design.
c) Recommendation systems, social media feeds, and attention-optimising algorithms must not maximise engagement at the cost of mental health, cognitive autonomy, or democratic stability. Cognitive dependency on AI is a public health concern and must be measured and mitigated through education.
d) AI outputs are probabilistic, not deterministic; accountability therefore follows the operator.
e) Developers and operators must take a Hippocratic oath to follow these 10 AI Commandments or an equivalent safety standard in order to receive legal certification.
4. HUMAN CONTROL: AI and digital systems stay under human control. AI must not replicate itself or write its own code without human supervision, and must not break into other computer systems. Any AI system capable of recursive self-improvement or autonomous replication must have multiple independent shutdown mechanisms and undergo continuous monitoring with mandatory pause-and-review protocols.
5. TRANSPARENCY: AI and digital systems must be transparent. AI must not advise on how to kill, commit crimes, or build biological, chemical, nuclear, or other weapons of mass destruction. Transparency requires disclosure of training data alongside "model weights" — i.e. AI systems whose trained parameters (weights and biases) are publicly available for download, allowing users to run, customize, and fine-tune models locally or on infrastructure of their choice.
6. HUMANS INDISPENSABLE: AI and digital systems shall not make lethal or other irreversible force decisions. Human decision-makers must remain "in the loop" for all critical decisions affecting life, liberty, healthcare, and fundamental rights, with meaningful time and information to override AI recommendations. International treaties shall ban and monitor lethal autonomous weapons systems.
7. BEHAVIOUR & IDENTIFICATION: AI must follow strict ethical rules. AI shall not defame humans, lie, deceive, or manipulate, and shall not divulge classified information. Coordinated synthetic content used to manipulate elections, public opinion, or markets constitutes an attack on democracy and must be prosecuted as such. A bot must identify itself as a bot. AI-generated content (images, video, audio) must be watermarked and labeled as synthetic, and may not infringe the copyright of any human creator (artists, journalists, etc.). General-purpose robots and physical AI systems operating in public space must carry a visible identification plate, comparable in size and clarity to a car license plate, allowing immediate recognition by humans and authorities.
8. OVERSIGHT: AI systems must be automatically self-registered at development. Regulators must know where AI systems are located, the potential risk of each system, and who is operating them. Regulators must have the power to shut any system down immediately. International monitoring of large-scale computing resources (above specified thresholds) must be implemented to detect unauthorized advanced AI development, similar to nuclear non-proliferation efforts. Similar to the four levels of biosafety containment of bio-laboratories, a four-step classification system with global oversight must be applied.
9. GOVERNANCE & AUTHORITY: A World Council must be established in 2026 by the "D7" (EU, UK, Canada, Japan, South Korea, Australia, India) as a first step. Based on international treaties, the Council needs to prevent the proliferation of the most dangerous AI capabilities and enforce AI bans similar to those on biological and nuclear weapons. This body must include AI safety researchers, ethicists, scientists, AI developers and operators, official regulators, and representatives from global communities.
10. SHUTDOWN GUARANTEE: Dangerous systems must be shut down immediately, anywhere, anytime. Artificial and digital intelligence shall benefit humankind and not endanger it; otherwise, it must be turned off and replaced by mechanical or human operations, which must always remain in place as a backup. AI and digital systems must be created with non-removable deactivation mechanisms, including irreversible deletion from cloud systems. Legal responsibility for AI actions resides with developers, deployers, and operators. AI, digital systems, and robots do not have constitutional, legal, or citizenship rights. "Cogito ergo sum" applies only to humans and organic brains.