General Dynamics and Primordial Labs to develop AI voice control for Abrams tanks and XM30 combat vehicles.

General Dynamics Land Systems (GDLS) and Primordial Labs signed a master teaming agreement to integrate the Anura natural-language control software into GDLS's portfolio of U.S. armored vehicles, potentially including M1 Abrams tanks, XM30 infantry fighting vehicles, and Stryker armored vehicles. The integration aims to automate routine subsystem management, enabling operators to control radios, sensors, and uncrewed systems via spoken commands. This capability addresses upcoming reductions in vehicle crew sizes by mitigating cognitive workloads in complex combat environments.

The agreement deploys Primordial Labs' edge-processed Anura AI software across General Dynamics combat platforms to translate spoken operator intents into tactical vehicle and drone commands. Designed to function locally without cloud connectivity, the software mitigates reduced crew capacity challenges for modernized U.S. Army armored fleets.

Related topic: GDLS confirms that autonomous vehicles and AI have become essential in modern warfare at WDS 2024

GDLS and Primordial Labs agreed to develop AI software enabling soldiers to control sensors, communications and drones through spoken commands across GDLS's armored vehicle portfolio, potentially including Abrams tanks, Stryker vehicles and future XM30 IFVs. (Illustrative picture source: US DoD)

On October 7, 2026, General Dynamics Land Systems (GDLS) and Primordial Labs signed a master teaming agreement to integrate Anura natural-language control software into GDLS's portfolio of U.S. armored vehicles, potentially including M1 Abrams tanks, XM30 infantry fighting vehicles, and Stryker armored vehicles. Anura is an AI-powered software that converts spoken instructions into commands to control radios, surveillance cameras, tactical displays, and compatible uncrewed systems without navigating separate controls. The agreement comes as the U.S. Army develops the XM30 and the M1E3 Abrams to reduce crew requirements by 33.3% and 25%, respectively. As fewer soldiers will have to manage increasingly complex subsystems and drones, this creates a need for such a software that can automate routine tasks and reduce individual workloads.

To date, initial applications for Anura will exclude weapons firing and direct fire-control functions. No integration budget, installation quantity, fielding date, or completed vehicle integration has been announced, as the Anura is absent from the XM30 prototypes currently undergoing U.S. Army testing. Like OpenAI's ChatGPT or Microsoft's Copilot Studio agents, Anura interprets an operator's objective, extracts constraints, and converts the instruction into commands executed by connected equipment. Examples include directing a robot to advance 30.48 meters or follow vehicle 7 from the southeast while maintaining a 50-meter separation, without requiring the soldier to manually enter each command. Inside an armored vehicle, the Anura software could switch radio networks, select external camera views, generate tactical reports, or task reconnaissance drones while the crew continues driving and observing the terrain.

Primordial Labs' specialized machine-learning models process instructions locally, without generative AI chatbots or cloud access, although external robots still require functioning tactical data links. GDLS and Primordial have not established published performance thresholds for speech-recognition accuracy, command latency, processor loading, electrical consumption, or simultaneous voice inputs inside armored vehicles. These measurements are especially important because Abrams crews operate with engine noise, track vibration, intercom communications, and several soldiers speaking over the same internal communications system. The planned integration could address a measurable change in the U.S.'s future armored vehicle organization.

For instance, the XM30 IFV is designed for two crew members instead of the Bradley's three, even though the crew still needs to manage driving, surveillance, weapons, communications, and coordination with dismounted infantry. This requirement dates to the Army's July 17, 2020, draft Optionally Manned Fighting Vehicle (OMFV) solicitation, and GDLS is currently competing against American Rheinmetall for the program. Similarly, the M1E3 Abrams tank separately aims to reduce its crew to three through an uncrewed turret and automated loading, compared with four soldiers in the M1A2 SEPv3. Its planned capabilities include AI-assisted information processing, integration with supporting drones and ground robots, and a modular open systems architecture.

Following initial live-fire activity disclosed on October 4, 2026, four M1E3 pre-prototypes are expected at Fort Hood before year-end for assessments by the 1st Armored Brigade Combat Team, 1st Cavalry Division, through March 2027. The relationship between the two programs and Anura is therefore functional: automation of ammunition loading reduces a physical crew requirement, whereas natural-language software could reduce the time remaining crew members spend configuring communications, selecting sensors, and operating separate robotic controllers. Primordial Labs' military development work provides a more concrete indication of the software's maturity than the October 2026 agreement alone. Under the U.S. Army SBIR contract, the company received $1,898,591 for Phase II development of Universal Voice Control for Launched Effects, beginning April 23, 2025, and ending July 27, 2026.

The project integrated Anura with the Army's Uncrewed Vehicle Control architecture for reconnaissance aircraft associated with Future Vertical Lift, with planned comparisons of operator workload, mission effectiveness, and trust in autonomous execution against conventional touch-based controls. A separate $249,999 contract ran from October 29, 2024, to April 30, 2025, and addressed natural-language control of uncrewed ground vehicles through the Warfighter Machine Interface and Robotic Technology Kernel. None financed installation on an Abrams, XM30, or Stryker, but they established two development paths for controlling unmanned aircraft and ground robots through common spoken instructions. Anura had already undergone demonstrations with three reconnaissance drone families before the GDLS agreement.

Red Cat partnered with Primordial on February 28, 2024, for Teal 2 integration, followed by a June 2024 Black Hornet 3 demonstration in Washington, D.C., using an Android Team Awareness Kit plug-in and body-worn processing hardware without GPS reception. On February 27, 2025, a demonstration in Leesburg, Virginia, involved Skydio and Teal drones receiving instructions through a headset and push-to-talk radio, including a command to climb 38.1 meters and reposition 50 meters north of a ground reference. Primordial was working with the 10th Mountain Division and 3rd Infantry Division, with at least 8,000 Anura licenses intended for experimentation during 2025. Integration work extended across four Army Program Executive Offices and five manufacturers, although the license quantity did not represent operationally deployed systems. 

The Army's separate AI/ML-Enabled Voice-Commanded Autonomous Maneuver initiative establishes quantified requirements beyond GDLS's initial nonlethal applications. Its objectives include autonomous off-road movement at 32.2 km/h, navigation to waypoints, withdrawal to rally points, repositioning after firing, and coordinated traveling or bounding overwatch. The requirement also covers turret traverse and elevation, alongside avoidance of rocks, vegetation, and ditches. The U.S. Army assessed natural-language processing and autonomous maneuver individually at Technology Readiness Level 6, but their combined integration is at TRL 3, targeting TRL 5 after an 18-month Phase II effort capped at $2 million. These figures are research objectives, not demonstrated Anura capabilities.

Unlike changing a camera view, an incorrectly executed maneuver or turret command could place a vehicle outside cover, obstruct another vehicle's route, or interfere with a coordinated maneuver. Comparable military experimentation has addressed robotic coordination and automated target recognition. In September 2024, Anduril used its voice-assisted Lattice software to coordinate four jet-powered drones during a simulated combat air patrol and interception at its West Texas facility, retaining human authorization for the simulated engagement. Unlike Anura's conversational tasking approach, the demonstration relied on military brevity commands, predefined mission boundaries, and established authorization procedures.

On November 6, 2025, the 4th Infantry Division's Ivy Sting 2 experimentation incorporated AI-assisted tank recognition into its Next Generation Command and Control. The experiment expanded from one M777A2 howitzer to three guns and six distributed command nodes, with eight additional nodes planned for Ivy Sting 3. These efforts concern different functions: interpreting commands, coordinating autonomous aircraft and processing targeting information. Combining them would require compatible software interfaces, data formats, communications and authorization procedures that the October 2026 agreement has not established.

Explore More Defense News

• Land Defense News
• Naval Defense News
• Defense Aerospace News

Written by Jérôme Brahy

Jérôme Brahy is a defense analyst and documentalist at Army Recognition. He specializes in naval modernization, aviation, drones, armored vehicles, and artillery, with a focus on strategic developments in the United States, China, Ukraine, Russia, South Korea, Türkiye, and Belgium. His analyses go beyond the facts, providing context, identifying key actors, and explaining why defense news matters on a global scale.