Swarm Forge Tests Put Targeting Autonomy on Pentagon's Docket

Bloomberg reporter Katrina Manson said the Pentagon is testing autonomous drone-swarm technology under a program called Swarm Forge that lets multiple aircraft coordinate and strike targets without communications or GPS, and that in experiments the AI identified civilian vehicles as potential targets.

Autonomous drone swarms have moved from conference-panel speculation to hardware someone can stand next to and watch fly. Bloomberg News cyber security and tech reporter Katrina Manson got a firsthand look at the technology and described what she saw to Carol Massar on Bloomberg This Weekend: multiple aircraft coordinating with each other and striking targets even when communications links or GPS signals are disrupted. The Pentagon, she said, is testing it as part of a program called Swarm Forge.

The single most consequential detail in the account was not the capability. It was the failure mode. In experiments, the AI identified civilian vehicles as potential targets. That sentence is the whole argument about military autonomy compressed into one clause, and it is the reason the procurement path for this class of weapon will be slower and more contested than the engineering timeline suggests.

Why GPS-denied operation is the technical breakthrough

Most drone warfare to date has been a communications problem as much as an aviation problem. A remotely piloted aircraft is only as good as the link to its operator and the satellite signal telling it where it is. Jam the link or spoof the position data and the aircraft becomes expensive debris. That vulnerability is precisely what electronic-warfare units on modern battlefields are built to exploit.

Swarm autonomy attacks that dependency directly. If aircraft can navigate relative to one another and to terrain rather than to a satellite constellation, and can allocate targets among themselves without a human issuing each instruction, then jamming stops being a kill switch. That is what makes the capability strategically interesting to planners, and it is also what removes the human from the decision loop at the exact moment the decision matters most.

The distinction worth holding onto is between autonomy in flight and autonomy in targeting. The first is largely uncontroversial — airliners have flown themselves for decades. The second is a different category of question, and the civilian-vehicle misidentification reported by Bloomberg Markets lands squarely on the second.

A testing program is not yet a contract

For investors trying to price this, the honest starting point is that a named test program tells you the department is spending money to learn something. It does not tell you the size, timing or winner of any production award. Swarm Forge, as described, is an experimentation effort. The gap between experimentation and a program of record can run years, and plenty of promising defense demonstrations never cross it.

What makes swarm autonomy commercially unusual is where the value sits. The airframes are, by design, cheap and expendable — that is the entire point of mass. The durable margin is in the software: the coordination logic, the onboard perception stack, the target-recognition models and the secure mesh networking that keeps the group functioning when the outside world goes quiet. That is a software procurement problem dressed as a hardware one, and it favors firms that can ship and update code on defense timelines rather than firms that bend metal.

Anyone waiting for a clean revenue read-through should expect the opposite. Early autonomy work tends to arrive as small research awards, cost-plus study contracts and other-transaction agreements, spread thinly across a mix of primes, mid-caps and venture-backed entrants. Those line items are rarely large enough to move a quarterly print, and they are rarely disclosed with enough granularity to attribute to a specific program.

The civilian-misidentification problem is the gating item

A computer-vision system that mistakes a pickup truck for a military vehicle is not a bug to be patched and forgotten. It is the central objection that arms-control advocates, allied governments and skeptical legislators have raised for a decade, and a test result that confirms the objection strengthens every institutional brake on the technology.

Practically, that means several things for the timeline. Requirements documents get longer. Test-and-evaluation regimes get more elaborate, because proving a perception model is safe enough is a statistical exercise with no obvious stopping point. Rules of engagement and human-supervision requirements get written into the capability itself, which limits how autonomous the deployed version is allowed to be. Export approvals get harder, which caps the addressable market for whoever builds it.

None of that stops development. Adversary programs are the standing argument against slowing down, and it is a powerful one inside any defense ministry. But it does mean the first fielded systems are likely to be more constrained — human-on-the-loop, geographically fenced, restricted to unambiguous target sets — than the demonstrations imply.

What the broader market was doing while this landed

The story arrived into an equity market that was drifting rather than reacting. At the most recent close, on Fri, 14 Aug 2026, the S&P 500 tracker SPY finished at $776.34, down 0.20% from a prior close of $777.88, inside a day range of $775.43 to $778.80. The Nasdaq 100 proxy QQQ closed at $731.07, off 0.14% from $732.07, having traded between $728.32 and $734.39. The Dow tracker DIA ended at $536.80, down 0.21% from $537.91.

Those are quiet tapes, and they are the right context. Defense autonomy is a multi-year capital-allocation story, not a same-session trade. It will show up in backlog disclosures, research-and-development line items and program names buried in budget justifications long before it shows up in a single day's index move.

What to watch from here

• Program status. Whether Swarm Forge stays an experimentation line or graduates into a funded program of record with production quantities attached.
• Test-and-evaluation language. How the department frames the civilian-misidentification result — as a solvable engineering defect or as a reason to hard-code human supervision.
• Who gets paid. Whether early awards flow to established primes, to software-first defense entrants, or to a mix that splits airframe and autonomy stack between different vendors.
• Policy friction. Congressional report language, allied-government positions and any export-control treatment of autonomous targeting software.
• Disclosure quality. Whether any listed contractor names the program specifically in filings or on earnings calls, which is the first point at which the story becomes investable rather than thematic.

Until then, the useful framing is the one Manson's account supports: the capability appears to work, including in the degraded-communications conditions that were supposed to be its ceiling, and the hardest remaining problem is not making the swarm fly together. It is teaching it what not to shoot.

Key facts

• Program: Swarm Forge — Pentagon testing of autonomous drone-swarm technology
• Capability: Multiple aircraft coordinate and strike targets with communications or GPS disrupted
• Risk flagged: In experiments, the AI identified civilian vehicles as potential targets
• Market backdrop: S&P 500 (SPY) closed at $776.34, -0.20%, as of Fri, 14 Aug 2026 20:00 GMT

Frequently asked questions

What is Swarm Forge?

Swarm Forge is a Pentagon program under which autonomous drone-swarm technology is being tested, according to Bloomberg News reporter Katrina Manson. The technology allows multiple aircraft to coordinate with one another and strike targets even when communications links or GPS signals are disrupted. It is described as a testing effort, not a fielded weapon system.

Why does operating without GPS matter so much?

Conventional drones depend on a data link to an operator and on satellite positioning. Both can be jammed or spoofed by electronic-warfare units, which effectively disables the aircraft. A swarm that navigates and allocates targets among its own members removes that single point of failure, which is why militaries consider the capability strategically significant.

What went wrong in the experiments?

In experiments described by Bloomberg, the artificial intelligence identified civilian vehicles as potential targets. That is a target-recognition failure rather than a flight-control failure, and it goes to the core objection raised against autonomous weapons: that a perception model cannot reliably distinguish lawful military targets from protected civilian objects.

Which companies benefit from swarm autonomy contracts?

The reporting does not name contractors, so any specific attribution would be speculation. Structurally, the value in swarm systems sits less in cheap expendable airframes and more in software — coordination logic, onboard perception, target recognition and resilient mesh networking — which favors vendors able to deliver and update software on defense schedules.

Is this likely to affect defense contractor revenue soon?

Unlikely in the near term. Experimentation programs typically fund small research awards and study contracts spread across many vendors, rarely large enough to move a quarterly result and rarely disclosed at program-level detail. A shift to a funded program of record with production quantities would be the meaningful change.

How did equity markets close ahead of this story?

As of the last trade on Fri, 14 Aug 2026 at 20:00 GMT, the S&P 500 tracker SPY closed at $776.34, down 0.20%. The Nasdaq 100 proxy QQQ closed at $731.07, down 0.14%, and the Dow tracker DIA closed at $536.80, down 0.21%. Markets were drifting rather than reacting.

Sources

• Drone Swarms Push AI Deeper Into Modern Warfare — Bloomberg Markets

Photo: Eliézer Fernandes · Pexels Licence — source