Model Superhuman Artificial Intelligence Safety and Humanity Protection Act

Draft legislative template; adapt to the constitutional and legal requirements of the adopting jurisdiction.

Section 1. Purpose

This Act protects human life, dignity, fundamental rights, democratic self-government, and the environment from harmful uses of artificial intelligence whose capabilities exceed those of humans. Such systems shall remain subject to effective human oversight and lawful public authority.

Section 2. Definitions

For purposes of this Act:

1. “Superhuman artificial intelligence” means an artificial intelligence system that substantially exceeds the capabilities of qualified human experts in either:
   • a broad range of consequential cognitive tasks; or
   • a particular domain in which its capabilities could cause serious or widespread harm.

Systems operating together shall be assessed according to their combined capabilities.

1. “Covered person” means any individual, corporation, public body, or other entity that develops, trains, possesses, deploys, controls, or materially modifies such a system.

2. “Harm against humanity” means:

   • Death, serious physical injury, or severe psychological injury;
   • Genocide, crimes against humanity, slavery, torture, or persecution;
   • Substantial violations of fundamental rights, including unlawful discrimination, coercion, or mass surveillance;
   • Destruction or serious disruption of essential infrastructure or services that endangers human life or public safety;
   • Severe or lasting environmental destruction;
   • Unlawful interference with elections, democratic institutions, or lawful human governance; or
   • Human extinction, widespread human incapacitation, or durable loss of meaningful human control over consequential decisions.

3. “Effective human control” means the practical ability of authorized, accountable persons to understand material risks, restrict a system’s authority, interrupt its operation, and safely shut it down.

Section 3. Prohibited Activities

1. No covered person shall develop, train, deploy, direct, or use superhuman artificial intelligence to cause, facilitate, or materially assist harm against humanity.

2. No covered person shall knowingly or recklessly permit such a system to engage in prohibited conduct where that person has the authority and practical ability to prevent it.

3. Prohibited activities include:

   • Creating or deploying biological, chemical, nuclear, or other weapons for unlawful use;
   • Selecting and attacking human targets without meaningful human control;
   • Conducting destructive cyberattacks or sabotaging critical infrastructure;
   • Coercing, deceiving, or manipulating people in ways that substantially impair their rights or autonomy;
   • Seizing governmental authority or essential resources without lawful authorization; and
   • Enabling unauthorized self-replication, resource acquisition, evasion of oversight, or interference with shutdown mechanisms where this creates a substantial risk of harm against humanity.

4. These prohibitions apply regardless of whether the activity is described as commercial, military, governmental, scientific, or experimental. An AI-generated instruction or decision does not excuse a violation.

Section 4. Mandatory Safety Requirements

Before deployment, and throughout operation, covered persons shall:

1. Conduct and document independent safety, security, and human-rights assessments proportionate to the system’s capabilities;
2. Demonstrate effective human control, including tested interruption and shutdown procedures;
3. Restrict access to weapons, critical infrastructure, sensitive data, and financial or computing resources according to demonstrated safety;
4. Maintain safeguards against misuse, unauthorized modification, and circumvention of restrictions;
5. Monitor for dangerous capabilities and materially changed behavior;
6. Retain auditable records while protecting privacy and confidential information; and
7. Immediately suspend affected operations when there is a credible indication of imminent serious harm.

Deployment shall not proceed when substantial risks of harm against humanity cannot be adequately controlled.

Section 5. Authorization and Oversight

1. An independent AI Safety Authority, established by this Act, shall license deployments presenting a substantial risk of widespread or catastrophic harm.
2. The Authority may require evaluations, inspect relevant records, impose operating conditions, and order restriction or suspension of unsafe systems.
3. Significant capability increases or material changes in use require renewed assessment.
4. Oversight shall include public accountability, protection of sensitive information, and timely judicial review.
5. Government systems are subject to this Act. Classified oversight procedures shall not waive its substantive protections.

Section 6. Incident Reporting and Emergency Action

Covered persons shall promptly report serious incidents, loss of control, and credible catastrophic risks to the Authority.

The Authority may order narrowly tailored emergency measures necessary to prevent imminent serious harm. Such orders shall be time-limited, documented, and subject to prompt judicial review.

Section 7. Liability and Remedies

1. Courts may issue injunctions, order corrective measures, and award compensation to persons injured by violations.
2. Civil penalties shall reflect the seriousness of the violation, culpability, financial benefit, and capacity to pay.
3. Knowing or reckless violations causing or creating a substantial risk of death, serious injury, or catastrophic harm shall constitute criminal offenses, with penalties established in the adopting jurisdiction’s criminal code.
4. Liability shall not be avoided merely by delegating decisions to an AI system.

Section 8. Protected Activities and Review

Good-faith safety research, incident reporting, and whistleblowing shall be protected when conducted with appropriate safeguards and without committing prohibited acts.

The legislature shall periodically review this Act as capabilities evolve. Regulatory changes shall not weaken the prohibitions in Section 3.

Section 9. Effective Date

The prohibitions in Section 3 take effect upon enactment. Other requirements take effect within 180 days, without limiting existing legal duties or emergency powers.