Texas-based Warhound was conceived by Anduril Industries founder Palmer Luckey and entrepreneur Ben Lamm, who started Colossal Biosciences, a company best known for its efforts to revive extinct species.

In a press release, Warhound chief executive David Spirk said the company aims to reduce the United States’ dependence on a “shrinking” European supply chain for working dogs.

“Our long-term goal is simple: healthier dogs, better-prepared teams, and more service members and first responders coming home safely,” Spirk said, adding that he cares deeply about “protecting people and helping the animals who support them.”

Before joining Warhound, Spirk held several defense-related positions, including senior counsel at Palantir Technologies and chief data officer at both the Defense Department and U.S. Special Operations Command, or SOCOM.

According to the announcement, Luckey and Lamm conceived the idea for Warhound years ago at the annual Moonshots Summit, a gathering where entrepreneurs discuss ambitious ideas.

They recognized two challenges with military working-dog programs: Most of the canines never complete training, and the U.S. relies heavily on a foreign supply chain. 

According to Warhound, overseas suppliers provide 80% of military working dogs. The company also says the United States must train roughly 200 dogs to field 100, in part because so many fail to complete training.

Warhound plans to combine the founders’ respective areas of expertise to address those problems.

“Colossal’s science will help produce healthier dogs with characteristics suited to demanding missions, while Anduril’s technology will give dogs and their handlers greater awareness and protection in the field,” the company said.

Warhound also said it will work with the American Humane Society to safeguard the animals’ health and welfare. The company said its goal is to reduce training washouts, limit the number of dogs entering intensive training unnecessarily and enable safer, more effective missions.

The goal is for AI to automatically analyze refueling operations from cameras mounted on the tankers and provide “hot wash” feedback to the tanker crews.

According to an Air Force Small Business Innovation Research solicitation, the service wants a system that can “leverage advanced computer vision, machine learning, and aviation (flight dynamics) analytics techniques to automatically extract operationally relevant performance metrics from aerial refueling events, quantify deviations from desired flight profiles, and generate actionable post-flight feedback to improve aircrew proficiency, reduce human-factor-driven mishaps, and enhance mission readiness.”

The application deadline is Nov. 25.

The SBIR is focused on the KC-46 tanker, which will feature a remote vision system that will allow operators to remotely operate the refueling boom via cameras and a 3D display. Though the system has been plagued by teething problems, the Air Force says most will soon be resolved.

The system would automatically track the tanker and refueling boom, as well as the receiving aircraft. It would measure a variety of data, including position, velocity, alignment and closure rate, and assess whether they deviate from nominal parameters.

The Air Force hopes an AI-based assessment system will mitigate the biases inherent in human assessment of aerial refueling, which is among the most demanding tasks for aircrew.

“Current aerial refueling training and post-flight assessment processes rely heavily on instructor observation, pilot self-assessment, manual video review, and subjective evaluation of performance,” the SBIR noted.

The system would create visual displays to improve the performance of pilots of the aircraft receiving fuel, such as “graphic AI overlays, heat maps, event timelines, and immersive or interactive debrief environments.” It would also have the potential to integrate with other training systems, including virtual and augmented reality, simulators, mission rehearsal systems, and other computer-based training tools.

The Air Force sees this as a prototype for future AI-based training systems.

“The desired end state is an operationally demonstrated capability ready to implement for at least one type of receiver aircraft and that can support transition into existing aviation training ecosystems and provide a scalable foundation for future AI-enabled aviation performance assessment across the Department of Defense,” the solicitation reads.

The orders are for the supply of the A2CS combat vehicles and the new-generation MBT tanks and are part of Italy’s wider military modernization program, the Italian financial daily added.

Rheinmetall declined to comment.

Leonardo CEO Lorenzo Mariani said he could not confirm whether the contract had been signed “as any communication on the matter must be coordinated with our clients.”

Leonardo and Rheinmetall launched their 50-50 joint venture in 2024.

“The programme is ready to enter production from our side. We are actively working on both the tank programme and the Future Combat System (FCS), and are progressing on both in close cooperation with Rheinmetall,” Mariani said in Rome.

Speaking on the sidelines of the annual meeting of the state-backed Fondo Italiano d’Investimento, he stressed the importance of tanks for the modernization of Italy’s defense, for exports and for the broader partnership with Rheinmetall.

Commander of I Corps Lt. Gen. Matt McFarlane told reporters at a media roundtable in Washington that the desire to see the platform deployed across the formation by the end of fiscal 2027 was dependent on what the service was able to provide.

“Scaling and speed up is based on resources available,” said McFarlane. “That’s not in my job jar.”

The Army’s NGC2 platform will permit technologies, weapons, sensors, devices and applications from across the force to co-exist on the same network, streamlining decision-making during combat environments.

As the service expands the new platform across the I Corps, which oversees troops across the Indo-Pacific, the Army will get a close-up look at how the system allows military personnel and machines to communicate with each other over vast geographical distances and in communication-contested environments.

Among its many responsibilities, the I Corps works with Security Assistance Group-Ukraine, a U.S.-led joint military command headquartered in Germany that provides weapons and training support to the Armed Forces of Ukraine.

I Corps uses SAG-U to digest lessons learned from the unique military dynamics of the war in Ukraine and determine how to apply them to other potential conflicts worldwide.

The Army recently awarded defense tech company Anduril Industries a contract worth up to $1.8 billion to continue developing NGC2’s common data baseline, or software foundation.

Anduril, along with Palantir, Raft and Striveworks, has been refining the technology to help the service move NGC2 from the experimental phase to fielding.

The resources McFarlane referred to include funds, which are currently up in the air as the Senate still works to pass the National Defense Authorization Act.

With Congress currently in recess, the timeline for approval of those funds is murky.

The Army’s proposed budget for fiscal 2027 includes a $904 million request for the research, development and scaling of NGC2.

“This enterprise licensing provides all assigned and attached personnel at any echelon within the specified commands the ability to gain MSS accounts within those domains they have authorized access,” the service said in last week’s announcement.

Maven is an AI-enabled software platform owned by the government but developed by the company Palantir. The system aggregates and analyzes data from several feeds like intelligence reports, drone video, satellite imagery, maps and more in support of real-time decision-making by commanders and analysts across services and allies.

For example, Defense Department officials say they used Maven to identify and bomb more than 2,000 distinct targets in Iran within the first 96 hours of Operation Epic Fury. Officials add that the system is used to process about 6 million pages of text per day for intelligence and operational workflows.

Ahead of a Maven training summit in August, the Marine Corps also explained that its demand for Maven has grown, noting that some 5,500 Marines used the system in the month prior. Additionally, the Marines replaced its manual process for filing Situational Reports with a Maven application.

Before the Oct. 1 agreement, the service’s license only allowed specific commands and organizations to use the AI-enabled battle management system and only through a Secret Internet Protocol Router Network, or SIPRNet. In that August 2025 rollout, the service provided dozens of units with access and training to the system.

The Marine Corps is encouraging commands to designate an MSS implementation lead for their respective organization. The service said that training materials are available on the secure website Intelink. It added that further guidance such as an MSS Implementation Plan will be published separately.

The F-22 completed airborne tactical command and in-flight control of a YFQ-44A Fury earlier this month, the Air Force announced.

The fighter jet’s pilot served as an airborne “quarterback” and gave high-level directives to an Anduril-made Fury CCA. The unmanned, semi-autonomous aircraft’s onboard autonomy software interpreted and executed the directives in real time, per the release, performing in-flight maneuvers and simulated combat tasks.

“This achievement demonstrates what is possible when we are aligned with industry on a shared imperative: delivering capability to the warfighter at unprecedented speed,” Secretary of the Air Force Troy Meink said in the announcement.

“We are driving urgency into every part of acquisition from rapid integration and flight learning to disciplined assessment, so we can turn promising technology into combat capability at the pace the threat demands,” he continued.

Personnel from the 412th Test Wing, 422nd Test and Evaluation Squadron, 309th Software Engineering Group, Experimental Operations Unit, F-22 System Program Office and the CCA Software Autonomy and Teaming Branch collaborated during the demonstration.

The service plans to use data from this event to inform software integration, developmental testing and further CCA concepts, the announcement says.

This test follows other milestones in the CCA development after the service announced its intention to have 500 combat-ready, semi-autonomous systems by 2032, with an overall fleet size of 1,000.

The Anduril-made Fury executed its first live-fire test by conducting an end-to-end, beyond-line-of-sight strike against a simulated target during a live-fire test at Edwards Air Force Base, California, in July.

Alongside Anduril’s CCA, the service has made strides with General Atomics’ FQ-42A Vengeance, which was delivered to Creech Air Force Base, Nevada, last month to continue regular test and evaluation operations.

Anduril has yet to announce which Air Force base will house Fury, but Creech airmen conducted an Agile Combat Employment exercise earlier using the defense company’s aircraft, allowing airmen to complete sorties and build foundational tactics and procedures that will mold how CCA are integrated into the force.

Gecko Robotics announced Tuesday that the Navy is increasing the value of its initial contract to up to $300 million — just six months after the company reached an initial agreement with the sea service.

“Gecko’s AI and robotic technology can save the U.S. Navy up to $8 million in repair costs per ship and return vessels to sea quicker by identifying structural issues up to 50 times faster than traditional methods — reducing inspections from months to days,” Gecko Robotics said in a release.

The initial contract began as a 5-year, $54 million indefinite delivery, indefinite quantity deal for Gecko to work on 18 ships assigned to the Navy’s U.S. Pacific Fleet.

The company uses drones, wall-climbing robots and fixed sensors to gather data on components, decks, welds and hulls.

That information, paired with AI tools, is used to identify current and potential structural issues that may remain hidden to the naked eye.

A key part of the new contract investment is Gecko Robotics’ Komodo machine system, which analyzes the health of deck coatings by scanning — eliminating the need to physically dig into them.

This allows the Navy to plan repairs and map out damaged areas before the ship enters dry dock, something that previously only occurred after the maintenance cycle began.

The Navy has consistently stated its goal of reaching 80% fleet readiness by 2027, despite criticisms of the service’s deployment capabilities.

The Government Accountability Office reported in December 2024 that readiness rates among the Navy’s amphibious warfare ships fell to 46% between 2011 and 2020, hindering Marine Corps deployment and training plans.

In August 2025, that rate reportedly dipped to just 41%, resulting in a more than five-month gap in Marine Expeditionary Unit deployments that year and further straining resources amid the Trump administration’s push to counter the illicit drug trade in Latin America and the Caribbean.

Chief of Naval Operations Adm. Daryl Caudle, who adopted the 80% goal from his predecessor Adm. Lisa Franchetti, laid out the major tenets for achieving the ambitious milestone.

“Achieving this requires shorter maintenance cycles, increased spare-parts availability, improved training pipelines and targeted upgrades across the fleet,” Caudle wrote for Military Times in December.

APKWS kits convert unguided 2.75-inch rockets into laser-guided munitions. All four services field them, and the weapon has become a cheaper alternative to air-to-air missiles for shooting down drones.

Congress still has to approve and fund the buy, the department said. BAE said it will increase production while the two sides work out a multiyear contract. Neither side announced the deal’s cost.

The framework also outlines a path to increase output to 35,000 kits a year for U.S. allies. The department said that added capacity will come online as partner nations formally request kits through the Foreign Military Sales program.

“By securing multi-year demand, this agreement keeps production lines running at capacity, provides our forces with interceptor inventory, and flips the cost calculus in the taxpayer’s favor,” Michael Duffey, the Pentagon’s undersecretary for acquisition and sustainment, said in the announcement.

BAE, which builds the kits in Hudson, New Hampshire, and Austin, Texas, said in February that it had delivered its 100,000th kit.

The 25,000 figure in this framework agreement matches capacity BAE has cited before. A company official said in 2023 that APKWS was in full-rate production with room to grow to 25,000 units a year.

“APKWS delivers proven, reliable and affordable precision at scale to counter evolving threats and ensure our warfighters maintain a decisive advantage,” said Tom Arseneault, president and CEO of BAE Systems Inc.

The Navy’s current five-year APKWS contract with BAE, awarded in August 2025, is worth up to $1.74 billion. It covers up to 55,000 kits for the Navy, the Army and foreign customers, with work expected to be completed by December 2031.

Among allied buyers, Saudi Arabia won State Department approval in July for a possible $1.96 billion package that includes up to 20,000 APKWS kits, split evenly between air-to-air and air-to-ground versions. Italy is also weighing a buy of 5,031 kits for counter-drone defense.

The APKWS agreement is the latest in a series of framework deals the Pentagon has signed this year to boost munitions production.

In January, Lockheed Martin agreed to more than triple production of the Patriot Advanced Capability-3 Missile Segment Enhancement, or PAC-3 MSE, interceptor.

In March, BAE and Lockheed agreed to quadruple production of seekers for the Terminal High Altitude Area Defense, or THAAD, interceptor. A separate THAAD framework signed in January led to a June contract worth more than $35 billion for the interceptors themselves.

According to the Sept. 29 announcement, the five winners of the Army’s xTechHumanoid competition include ACES Group, IHMC Robotics, Hydroplane Limited, Ras Labs and Velocity Explorations.

Alex Miller, the Army’s chief technology officer, explained that through the program, the service tried to identify “the problems, mission needs, and opportunities” humanoid robots could best be used.

“xTech is our method of surveying the industry landscape and fast-tracking promising technology rather than falling victim to years-long studies,” Miller said.

“Humanoid robots are part of a broader autonomy push to increase the depth and breadth of robotic platforms and components industry can provide for a range of missions,” he added.

The expectation

For the competition, Army officials reviewed each company’s white paper proposal for its humanoid robot. They evaluated the concepts based on several criteria, including military utility, technical merit and technological maturity.

Of the 103 submissions, the Army selected 10 finalists, each of which received $25,000. The service then identified five winners through a live experimentation event.

According to the 2025 competition announcement, the event tested the robots on a series of collective tasks “in the context of a military small unit mission conducted in a combat situation.”

During the event, officials also reviewed the baseline system capabilities and advanced subsystems.

Examples of the former include checkpoint security, gap crossing or obstacle clearing and firefighting. Whereas the latter included technologies like thermal sensors, battery life and autonomy.

The reality

However, Robert Griffin, a senior researcher at IHMC Robotics, told Military Times that the competition’s individual requirements exceeded what current technology could achieve.

“The individual things that were outlined were, in any one element, beyond what is possible given the current state of the art,” Griffin said.

“For example, you don’t have a robot that can run 28 miles an hour while carrying 50 pounds and operating for 24 hours,” he said. “It’s not a thing that exists in the world.”

Because the Army’s “actual full desirements were beyond what can currently be done,” Griffin said, developers had to prioritize different elements of their robots’ capabilities for the demonstrations.

Griffin’s team focused on baseline capabilities.

“The big thing for us was to have the robot show up at the competition, move around in the world, unsupported, at a reasonable pace, and then respond to commands,” he said.

Video of the competition shows IHMC’s robot “Alex” following voice commands directing it to walk through a field and then, under the control of a human with a remote, through the woods and across gravel.

Alex also grasps a small shovel, begins to dig and attempts to pour dirt into a bucket. Although it doesn’t succeed, it demonstrates the movement. And finally, Alex collapses and has to be transported out on a gurney.

Testing subsystems

Other competitors focused on subsystems.

Hydroplane and ACES Group competed with their batteries, while Velocity Explorations entered software that enabled its robot to listen to radio traffic between people and understand their movements and tasks. Using that information to develop an understanding of the broader battlespace, the platform could then determine how to act.

Ras Labs tested highly sensitive fingertip tactile sensors that allowed a robot to handle delicate objects.

While multiple companies said they participated because it was a chance to test their products, Dr. Anita Sengupta, Hydroplane’s founder and chief executive, explained that her company has traditionally done work for large structures and aircraft, but recently started working on smaller and portable systems.

“We saw a good overlap between developing a power system for a humanoid, whether it’s a person or a robot, as well as the drone use case, so it just seemed like an interesting use case for us to take our technology to,” she said.

For the competition, her company adapted a 10-kilowatt system originally designed for the Marine Corps in 2024, but made it smaller because the humanoid only needed a few kilowatts. The device utilizes a fuel cell similar to what a generator uses, but instead of using diesel fuel, it uses hydrogen. Sengupta explained that the benefits are that it has about three times the energy density and a higher conversion efficiency.

Although Hydroplane didn’t directly power a humanoid robot with its fuel cell, the Army tested it to see if it could power a humanoid robot. In turn, Hydroplane helped establish those standards, with the intent to eventually collaborate with a humanoid robot team.

The future

In the recent announcement, the Army said the competition revealed a clear picture of “both the progress commercial industry has made in humanoid robotics and the technical challenges that remain.” The findings, the service said, will help it direct research toward technologies that offer the greatest return for warfighters.

But while humanoid robots might eventually take on more complex military tasks, Griffin said the industry is still a long way from developing machines like Arnold Schwarzenegger’s T-800 from The Terminator franchise.

“I don’t think that any person who has a modicum of ethics is going to say that we need to be having these robots making life or death decisions,” he said, noting how easily current AI models can be confused or manipulated.

Instead, Griffin said the future of humanoid robotics will depend in part on what investors want the machines to do. Industrial companies, for example, may see opportunities to automate tasks typically performed by humans.

“Our goal is to be able to make robots that remove people from having to do harmful tasks,” he said about his company, pointing to scenarios such as first responders entering burning buildings and nuclear cleanup.

“Hivemind Enterprise handled the autonomy basics, so [Honeywell engineers] could spend its time on its own detection and tracking model,” said Tom Schaefer, vice president of Hivemind Enterprise at Shield AI. “We built Hivemind so partners can work this way.”

The companies announced their collaboration in July along with a memorandum of understanding, stating their goal was to help “defense operators move from demonstration to deployment faster… without having to rebuild confidence in each new system every time.”

Shield AI explained the Hivemind software provides “core capabilities such as route planning, mission management, and communication between aircraft,” so companies like Honeywell could integrate “their own AI models, sensors, and mission capabilities and [advance] them from simulation to flight.”

According to the statement, Honeywell’s autonomy team began training on Hivemind in late April and by mid-June, they flew a BFD-SE8 multirotor aircraft autonomously for the first time using the newly developed autonomy stack.

The team held the official demonstration on July 8 in Plymouth, Minnesota, where the aircraft took off on its own, followed a planned route, re-routed around a no-fly zone, and returned to base.

Afterward, the company said, the team immediately began working on the second demonstration. For three months, they trained and deployed their own target detection and tracking model.

On Oct. 8, the team hosted another demonstration in which a scout aircraft used the AI model to detect, identify, and track a target. It then passed the information to another aircraft, which navigated autonomously to the target, delivered a payload, and then returned to base with the scout.

Todd Giles, Honeywell Aerospace’s senior vice president of engineering and chief technology officer, credited “open architectures” — or the use of public standards for software development — for “changing the pace of innovation in aerospace” and his team’s accomplishment.

“By combining our expertise in avionics, navigation, sensing, and autonomy, we are helping create a scalable foundation for the next generation of autonomous aircraft and mission systems,” Giles said.