S. 4707: Responsible Artificial Intelligence Defense Act of 2026

This bill would set new Department of Defense rules for how the military develops, reviews, tests, and uses autonomous weapons and artificial intelligence systems.

Overall policy

The bill says the Defense Department should use autonomy and AI as much as practical, but only with continuous, careful human oversight. It also says these systems must be used in ways that comply with the law of war, treaties, weapons safety rules, rules of engagement, and protections for the privacy and civil liberties of U.S. persons.

What the Defense Department would have to do

• Make sure people remain responsible for the development, deployment, and use of autonomous weapons and AI capabilities.
• Take steps to improve accuracy in these systems.
• Ensure the technology and how it is developed and used are understood well enough for oversight.
• Subject systems to regular testing, information security checks, and lifecycle reviews to make sure they stay safe, secure, and effective.
• Design systems so operators can detect unexpected behavior and shut systems down or disable them if needed.

Review requirements before development and deployment

Before a system moves into prototyping or formal development, the Under Secretary of Defense for Research and Engineering would have to verify that:

• human operators can exercise an appropriate level of judgment over use of force;
• the system is designed to act within intended time, place, and operational limits, or else stop and seek more operator input;
• the design considers risks to unintended targets;
• the system includes safety, anti-tamper, and cybersecurity protections;
• there are plans for verification, validation, testing, and evaluation under realistic conditions, including possible enemy actions; and
• a preliminary legal review has been completed with the Department of Defense General Counsel.

Before fielding a system, additional review would be required to confirm that training, tactics, human-machine interfaces, safety protections, and monitoring procedures are in place, and that an updated legal review has been completed.

Testing and follow-up reviews

The bill would require rigorous testing throughout a system’s life cycle, including:

• hardware and software verification and validation;
• realistic operational testing against adaptive adversaries;
• cyber testing where relevant;
• testing for new or changed operating modes;
• post-deployment monitoring to catch changes that could create failures or unintended behavior; and
• repeat reviews every three years, with similar system variants treated as already verified if they are substantially the same.

Privacy reviews for some AI systems

If an AI capability uses sensitive personal data protected by the Privacy Act or HIPAA-related rules, the Department would have to conduct a privacy impact assessment.

Exceptions

Several types of systems would be excluded from these requirements, including:

• certain operator-supervised systems used to intercept time-critical or saturation attacks;
• certain operator-supervised systems defending unmanned vehicles or vessels;
• non-lethal cyberspace capabilities;
• unarmed platforms;
• unguided munitions;
• manually guided munitions;
• mines and unexploded ordnance; and
• autonomous or semi-autonomous systems that are not weapon systems.

The bill also creates a narrow process for some systems that are not continuously monitored by a human operator: they could be used only if the Secretary of Defense determines they are safer and more reliable than alternatives, notifies Congress, and no defense committee objects within 30 days.

Prohibited uses

The bill would prohibit the Defense Department from using AI or autonomy to:

• decide whether to launch a nuclear weapon;
• monitor, track, profile, or target people reasonably believed to be in the United States without a warrant based on probable cause, or otherwise collect or analyze information in ways not permitted by the Constitution; or
• use lethal force through autonomous weapon systems without appropriate human judgment.

Working group and reporting

The bill would create an “Autonomy and Artificial Intelligence Working Group” to advise senior Pentagon officials, help decide whether systems need higher-level approval, and develop safety standards for evaluating these capabilities. The Secretary of Defense would also have to submit annual reports to Congress through January 31, 2037.

Relevant Companies

• None found