The takeaway: The US administration is pressing the Pentagon to reopen the rules on weapons that pick their own targets — not scrapping the human-control requirement, but reconsidering it, just as Congress splits over the same question and the hardware starts turning up on real battlefields.

The United States is quietly reopening one of the hardest questions in modern warfare: how much of the decision to kill can be handed to a machine.

According to a Washington Post feature published on 20 July 2026 ("How armed robots could become the military's weapon of choice"), the administration is pressing the Pentagon to revisit the policies that govern weapons capable of selecting and engaging targets autonomously. Because that report sits behind a paywall, we verified the core of it against accessible reporting — including the International Business Times, which independently reports that the administration directed the Pentagon to revisit its autonomous-weapons rules and that the President signed a national security memorandum directing officials to account for advances in AI. So the direction of travel is real. What remains ambiguous is how far the line actually moves.

What the "line" is — and what's being reconsidered

The line in question is DoD Directive 3000.09, which requires that autonomous and semi-autonomous weapon systems allow commanders and operators to "exercise appropriate levels of human judgment over the use of force." In the wider debate it is often shorthanded as "meaningful human control" — a phrase the Pentagon itself has pointedly avoided. Crucially, it has never had a crisp definition — a gap that both restrainers and accelerators are now trying to fill in opposite directions.

This is where the story gets its tension. Just two days before the Post feature, we covered a congressional effort pointing the other way — the Human Authority over Autonomous Weapons Act, which seeks to harden the requirement that a human authorize any lethal engagement. Meanwhile, the Senate Armed Services Committee's FY2027 National Defense Authorization Act framework goes the other way: per the Arms Control Association, it largely endorses lethal autonomous weapons and urges the department to "maximize uses" of them, while adding guardrails — human supervision, intervention and termination methods, fail-safes, and a categorical ban on autonomous systems deciding to launch a nuclear weapon.

So the honest summary is not "America is unleashing killer robots." It is messier and more revealing: the legislative branch is split, the executive branch is leaning forward, and nobody has yet agreed what "human control" concretely means when the software is fast and the target is fleeting.

The hardware is already here

What makes this more than a policy-paper fight is that the machines exist. Foundation Future Industries, a San Francisco startup founded in April 2024 and led by former fintech founder Sankaet Pathak, has secured a reported $24 million in research contracts across the Army, Navy, and Air Force (The Next Web). In May it released footage of its Phantom humanoid — a machine with an LLM-driven autonomy stack — loading a mortar. Two units were sent to Ukraine in February 2026 for logistics and reconnaissance, described as among the first humanoid robots deployed to a combat theater.

Foundation says human operators retain final authority over lethal decisions today — the human-in-the-loop posture Directive 3000.09 requires. But its own framing is that autonomy should reduce teleoperation over time, and that in some time-critical situations robots may need to act on their own. That is precisely the seam the policy debate is worrying at.

Even the suppliers are drawing lines

One of the more telling signals comes from the AI industry itself. Per IBTimes, Anthropic CEO Dario Amodei reportedly sought assurances from the Pentagon this spring that the company's Claude model would not be used to power fully autonomous weapons, citing reliability concerns. (We were unable to locate a verbatim quote from Amodei at an accessible source, so we characterize this as reported, not confirmed word-for-word.) OpenAI, for its part, prohibits use of its systems in applications involving the use of force without appropriate human authorization.

That a frontier lab building the underlying models is trying to fence off "fully autonomous" is worth sitting with. It suggests the people closest to the technology are not confident it is reliable enough to trust with the final call — even as the demand signal from the Pentagon grows. Joint Chiefs Chairman Gen. Dan Caine said in April that autonomous weapons will be a "key and essential part of everything we do" (Defense One).

The bottom line

Nothing here says a machine has been cleared to decide, unsupervised, to take a human life. What has changed is the posture: the administration wants the rulebook reopened, the hardware is on real battlefields, and the definition of "human control" is up for grabs at exactly the moment it matters most. Congress is pulling one way, the Pentagon another, and the robots aren't waiting for either.