The HALO Act of 2026 establishes mandatory design, safety, review, and testing requirements for the Department of Defense's use of autonomous and semi-autonomous weapon systems utilizing artificial intelligence, while also prohibiting specific high-risk AI applications.
Adam Schiff
Senator
CA
The HALO Act of 2026 establishes comprehensive design, safety, and rigorous testing requirements for the Department of Defense's use of autonomous and semi-autonomous weapon systems that utilize artificial intelligence. It mandates clear human responsibility and ultimate command control over the use of force, while strictly prohibiting the use of covered AI in ways that violate constitutional rights or civil liberties. The bill also institutes advanced review and approval processes before developing and fielding these systems, alongside mandatory reporting to Congress on compliance and use cases.
Human Authority in Lethal Operations Act of 2026 Sets Strict Human-in-the-Loop Rules for AI Weapons: Rollout Begins Within 180 Days.
The HALO Act of 2026 is essentially a "safety manual" for the future of warfare, designed to ensure that even as technology gets smarter, a human being is always the one pulling the trigger. Within 180 days of becoming law, the Department of Defense (DoD) must implement a strict chain of command where a "designated commander" is legally and ethically responsible for every action an AI-enabled weapon takes. This means if a system identifies and engages a target, there must be a clear paper trail showing that a person—not an algorithm—had the final say and the ultimate discretion over that use of force.
Putting a Face on the Algorithm
To keep things from getting too "sci-fi," the bill requires that every autonomous or semi-autonomous system be designed so that it’s easy for a human operator to understand. Think of it like a car’s dashboard: the system has to provide real-time feedback on why it’s making certain suggestions and what data it's looking at (Section 3). For a soldier on the ground or a technician in a command center, this means they can’t just "set it and forget it." They are required to undergo quarterly training to ensure they actually understand the AI’s limitations. If the Wi-Fi goes down or communication is lost, the bill mandates that unmanned platforms cannot autonomously pick new targets on their own—they have to wait for a human to get back online (Section 3).
The "No-Go" Zone for AI
The bill doesn't just regulate weapons; it sets hard boundaries on how the DoD can use AI against people in general. Section 7 explicitly forbids using AI to track or profile Americans based on their religion, race, or how they exercise their right to free speech. For example, the government can't use an AI tool to monitor your location via data brokers without a warrant, nor can it use AI to try and "read" your emotions. This is a big win for civil liberties, ensuring that the same tech used on the battlefield isn't turned inward to monitor domestic political activity or private lives.
The Fine Print and the Loopholes
While the bill is tough on lethal weapons, it leaves some doors open. Section 10 exempts "cyberspace capabilities" and "unarmed platforms" from many of these strict reviews. This means AI used for hacking or surveillance drones that don't carry missiles might not face the same level of scrutiny. Additionally, while the bill requires a human to be in the loop for nuclear decisions, it allows for "time-critical" defensive systems—like those that shoot down incoming rockets—to act autonomously if there isn't enough time for a human to react. The challenge here is the "vague" definition of what counts as a "densely populated area," which determines whether some of these automated defensive systems need extra layers of approval before they are deployed near civilian centers.