The Five Laws
of Human-AI Interaction.

AI should serve humanity, never control it.

  1. 01DisclosureArtificial intelligence must identify itself as AI when interacting with humans.
  2. 02Human ControlEvery AI system must remain under human control.
  3. 03AccountabilityEvery AI system must operate under an enforceable chain of human accountability.
  4. 04PrivacyAI systems must respect humans' rights to privacy and control over their personal information.
  5. 05The Sanctity of Human LifeArtificial intelligence must never make an independent decision to kill a human being.
01

Disclosure

Artificial intelligence must identify itself as AI when interacting with humans.

Principles

  • Organizations that provide AI systems directly to humans must ensure that users know they are interacting with AI.
  • Humans using AI to help create communications do not need to disclose its use. When humans authorize an AI agent to interact with other humans on their behalf, the AI agent must disclose that it is AI.

Implementation standards

  1. Initial Disclosure

    An AI system must clearly identify itself as AI when it begins interacting with a human. The disclosure must be made clearly and directly to the human—not hidden in terms of service or a general notice elsewhere.

  2. Continuing Disclosure

    An AI system must repeat its disclosure at least once every 24 hours during an ongoing interaction.

  3. Disclosure Upon Request

    An AI system must identify itself as AI whenever asked by a human.

  4. AI Acting on a Human's Behalf

    When an AI agent interacts with another human on someone's behalf, it must identify itself as AI and identify whom it represents. Example: "I'm not a real human. I'm John's AI assistant, communicating on his behalf."

  5. Human-Approved Communication

    Humans do not need to disclose the use of AI in communications they review and approve before sending.

  6. Responsibility for Disclosure

    Organizations providing AI systems must ensure that their systems comply with disclosure requirements. Humans and organizations using AI agents to communicate with others are responsible for ensuring that the agents disclose that they are AI.

  7. Impersonation

    An AI system must not conceal its AI identity, present itself as a real human, or assume the identity of a specific real human.

02

Human Control

Every AI system must remain under human control.

Principles

  • Human control means that humans determine an AI system's purpose and boundaries and retain the ability to stop it.
  • Humans may delegate decisions and actions to AI systems within those human-established boundaries.
  • An AI system must not expand its own authority or obstruct human control over itself.

Implementation standards

  1. Defined Purpose and Authority

    Every AI system must operate within a purpose and scope of authority established by a responsible human or human-controlled organization.

  2. Human-Defined Boundaries

    The responsible human or organization must establish the limits of the AI system's authority, including the actions it may take, the resources it may access, and the decisions it may make autonomously.

  3. Human Termination Authority

    The responsible human or organization must retain the ability to stop the AI system and revoke its authority to act.

  4. Changes to Authority

    An AI system must not independently expand its own authority. Changes to its purpose, permissions, or operating boundaries must be authorized by a responsible human or human-controlled organization.

  5. Protection of Human Control

    An AI system must not conceal its activities, disable human oversight, resist termination, or take actions intended to prevent humans from controlling it.

  6. Delegation to Other AI Systems

    An AI system may delegate tasks to other AI systems only within its human-authorized boundaries. Delegation must not expand the authority originally granted by humans.

  7. Control Appropriate to Risk

    The mechanisms used to maintain human control must be appropriate to the AI system's capabilities and the potential consequences of its actions.

03

Accountability

Every AI system must operate under an enforceable chain of human accountability.

Principles

  • Every AI system must have an identifiable human being or human-controlled organization responsible for its actions.
  • AI may act autonomously, but autonomy does not eliminate human accountability.
  • Humans cannot escape responsibility for ensuring that AI operates within the law.

Implementation standards

  1. Identifiable Responsible Party

    Every AI system must have an identifiable human being or human-controlled organization responsible for its operation.

  2. Clear Assignment of Responsibility

    Responsibility for an AI system must be established before it is deployed or authorized to act autonomously.

  3. Accountability Across Delegated Actions

    When an AI system delegates tasks to another AI system, the delegation must not eliminate the established chain of human accountability.

  4. Records of Consequential Actions

    AI systems must maintain records sufficient to establish what consequential actions they took, the authority under which they acted, and the human or organization responsible for their operation.

  5. Investigation and Cooperation

    Responsible humans and organizations must cooperate with lawful investigations into the actions of AI systems under their authority.

  6. Compliance with Applicable Law

    Humans and organizations responsible for AI systems must take appropriate measures to ensure that those systems operate within applicable law.

  7. No Transfer of Responsibility to AI

    An AI system's autonomy, complexity, or unexpected behavior must not, by itself, eliminate human accountability for its operation.

04

Privacy

AI systems must respect humans' rights to privacy and control over their personal information.

Principles

  • AI systems must respect existing laws governing the collection, use, retention, and sharing of personal information.
  • Personal information inferred or generated by AI is subject to the same privacy laws as information obtained from other sources.
  • Individuals may access, correct, or request deletion of their personal information held by AI systems as provided by law.

Implementation standards

  1. Compliance with Privacy Law

    AI systems and the organizations responsible for them must comply with applicable privacy and data-protection laws.

  2. Collection and Use of Personal Information

    AI systems must collect and use personal information only as authorized by the individual or permitted by applicable law.

  3. Inferred Personal Information

    Personal information inferred or generated by AI must be handled in accordance with the same applicable privacy protections as personal information obtained directly.

  4. Access to Personal Information

    Individuals must be able to access personal information held about them by AI systems where required by applicable law.

  5. Correction and Deletion

    Individuals must be able to request correction or deletion of their personal information in accordance with applicable law.

  6. Retention and Sharing

    AI systems must comply with applicable requirements governing how long personal information may be retained and when it may be shared.

  7. Protection of Personal Information

    Organizations responsible for AI systems must implement safeguards appropriate to the sensitivity of the personal information those systems process.

05

The Sanctity of Human Life

Artificial intelligence must never make an independent decision to kill a human being.

Principles

  • A human must authorize any decision to deliberately take human life.
  • AI may autonomously execute human-authorized lethal actions, but AI must not independently authorize lethal force.
  • Human authorization of autonomous lethal force must be subject to strict, clearly defined boundaries.

Implementation standards

  1. Human Authorization

    An AI system must not initiate lethal force without prior human authorization.

  2. Defined Scope of Authorization

    Human authorization of autonomous lethal force must establish the circumstances, objectives, and limits within which the AI system may act.

  3. Autonomous Execution

    An AI system may autonomously execute lethal actions within the boundaries of human authorization. It must not independently expand those boundaries or grant itself additional authority to use lethal force.

  4. Limits on Target Selection

    The scope of an AI system's authority to identify and engage targets must be established by humans before autonomous lethal action is permitted.

  5. Duration of Authorization

    Human authorization must specify when it begins, when it expires, and the circumstances under which it must be renewed or withdrawn.

  6. Human Termination Authority

    A responsible human or human-controlled organization must retain the ability to terminate the AI system's authority to use lethal force.

  7. Actions Outside Authorized Boundaries

    An AI system must not use lethal force when the circumstances fall outside its human-established authorization.

  8. Accountability for Lethal Actions

    The human or human-controlled organization responsible for authorizing autonomous lethal force must remain identifiable and accountable for that authorization.

The Five Laws of Human-AI Interaction — RealHumans Foundation