DARPA and the U.S. Air Force published an announcement on July 16 spelling out the project name in plain language: Viper Experimentation and Next-gen Operations Model (VENOM). The core move is installing a hardware-plus-software suite called the VAK (VENOM Autonomy Kit) on a modified F-16, letting AI agents fly the aircraft autonomously under real sensors, real weapon flight models, real dynamics, and real communications systems.
A pilot still sits in the cockpit — but control between the pilot and the aircraft can be swapped via a flip of a switch. This is the first time AI has performed autonomous control on an "operationally active fighter jet," rather than running simulations on an experimental platform like the X-62A VISTA.
Key program architecture and funding details:
- Program span: VENOM is jointly led by DARPA and the U.S. Air Force, housed under the AIR (Artificial Intelligence Reinforcements) program, whose parent is ACE (Air Combat Evolution)
- Investment: DARPA has put over $132 million into AIR (3 years plus VENOM kit implementation), running in parallel with thousands of hours of simulation and modeling with AI agents
- Location: 96th Test Wing and 53rd Wing at Eglin Air Force Base, Florida
- Test phases: Initial flight testing is complete → next steps in four stages: basic fighter maneuvers → 1v1 within visual range → 1v1 beyond visual range → 2v2 and 4v4 formation engagements, eventually adding electronic warfare scenarios
- Transition path: After AIR concludes, the AI agents validated in VENOM will transfer to the CCA (Collaborative Combat Aircraft) program — the U.S. military's fast-moving effort to deploy unmanned collaborative combat aircraft
- U.S. Air Force DVIDSHUB official release: https://www.dvidshub.net/news/570088/venom-program-progresses-piloted-flights-autonomy-tests
- DARPA's own announcement: "Historic VENOM milestone demonstrates scalable AI development capabilities for the operational fleet"
- AIAA Aerospace America in-depth interview with Brig. Gen. Valpiani: https://aerospaceamerica.aiaa.org/new-test-campaign-will-put-ai-pilots-in-most-realistic-flight-conditions-yet-darpa-official-says/
The VENOM team is led by DARPA program manager Brig. Gen. James "Fangs" Valpiani (whose term ends this month), with the baton passing to Lt. Col. Patrick "Dice" Highland, and deputy PM Terry Wilson, PhD (AI R&D and transition lead at AFRL ACT3) continuing. Organizationally, this is one of very few DARPA AI programs placing a general officer at the front-line PM position.
Situating this in the 2026 "physical AI / embodied intelligence" macro narrative, several concrete signals stand out:
1. A completely different engineering dimension from the X-62A VISTA — previous AI fighter programs relied on "perfect information + simulation substituting for real sensors." VENOM is the first to subject the AI pilot to "the same information incompleteness as a human pilot." In DARPA PM Valpiani's own words: "This really is a sea change. All previous AI autonomous fighter research got some kind of white card or some kind of simulation in place of the real human experience." No white card this time.
2. "Rack-level software, no changes to the aircraft's core code" is VENOM's real engineering philosophy — the AI kit connects to the F-16's existing flight controls via a novel interface, allowing the aircraft to present externally as an ordinary operational aircraft. This path is strikingly similar to Anthropic running AI on AMD GPUs "without changing hardware." The optimal way to add AI to physical systems isn't replacement — it's layering it on as a detachable / pluggable layer.
3. From single AI → formation AI is DARPA's next phase — DARPA is already training for 4v4 formations beyond VENOM (Valpiani: "perhaps one to two years to become proficient in as a new pilot"). This pushes the minimum tier of human-machine teaming from "AI copilot + 1 pilot" to "AI formation + 1 human commander." The path to CCA is therefore not "unmanned replacing manned," but "manned commanding unmanned formations."
4. The defense vs. civilian split in embodied intelligence is becoming clear — over the past six months, the embodied intelligence mainline has been industrial (UBTech, Tesla Optimus, Figure), home (1X Neo), and service robotics; but the defense line is being pushed to "the eve of combat" in parallel by China (Unitree's IPO prospectus, NUDT's new results) and the U.S. (this DARPA wave).
Full announcements: