Concerned about how artificial intelligence might be used to generate target lists or operational plans, lawmakers want to expand limits on autonomous weapons to address mission planning and target selection. The House Armed Services Committee’s version of the 2027 National Defense Authorization bill would direct the Pentagon to revise Defense Directive 3000.09 to cover mission planning activities.
If approved, it would mark a “big change,” said Bryan Clark of the Hudson Institute. The measure is among several moving through Congress that seek to address emerging AI and related capabilities, including one focused on orbital data centers, and another that would establish a new combatant command to oversee autonomous and robotic systems.
Under the plan inserted into the House markup of the bill, the Pentagon would have to establish or revise guidelines for “artificial intelligence-enabled systems intended to support, recommend, or materially influence operational decisions associated with the employment of force, including systems used for operational planning, target development, weaponeering, or engagement recommendation,” according to the HASC subcommittee on cyber, information technologies, and innovation.
A 2025 Congressional Research Services Report, “U.S. Policy on Lethal Autonomous Weapon Systems,” examined Directive 3000.09, noting that the policy document focused on how human operators might make target selection and engagement decisions using an autonomous weapon system. The directive, first published in 2012, was last updated in 2023.
With Operation Epic Fury only just concluded, the question of how those decisions are made are again under scrutiny.
“The [department] has been using a lot of AI-decision enabled support systems in the war in Iran for strike planning, for example, and what this is saying is that you now need to have some kind of governance around those too,” said Clark, who directs Hudson’s Center for Defense Concepts and Technology. Planners are no longer “proposing a kill chain to go blow something up, which is pretty simple,” Clark said. “Now [they’re] asking it to decide what are the right things to attack those targets and then what are the things I should use to attack those targets and then build me some kill chains to make that happen.”
The issues are thorny, he said, because conventional processes involve numerous planners with a wide range of expertise, an inherently complex group with built-in checks and balances. AI, by contrast, may take those factors into account, but understanding how a decision is made is almost impossible—a figurative black box, Clark said.
Retired Air Force Lt. Gen. Bill Bender, a former Air Force Chief Information Officer, sees value in Congressional oversight as AI technology evolves.
Using AI in mission and target planning systems “begs for legislative oversight, at a minimum,” Bender said. The Pentagon and Congress should get together to develop a measured, thoughtful approach on how to implement AI tools, while ensuring that controls do not unnecessarily constrain AI, leading to missed opportunities.
“The challenge will be loosening the restrictions in a timely fashion as better AI enablement has iteratively arrived upon improvements,” Bender said.
