Ukraine, in the fifth year of fighting a full-scale Russian invasion, is already using AI for a plethora of battlefield functions, from flying drones at targets to helping plan combat operations and crunching data on Russian missile attacks.

“AI will form a new paradigm of warfare. It’s already actively doing so,” Danylo Tsvok, the head of the defense ministry’s AI center, told Reuters.

He predicted AI systems would eventually be unified into a single network overseeing the battlefield, leading to a “war of operating systems” with Russia in the next three to five years, if the conflict continues.

“The system that possesses more data and better understands that data, proposes solutions — that system will gain the advantage over the other," he said.

The center was founded in March as Defense Minister Mykhailo Fedorov seeks to put AI and data-driven decision-making at the heart of Ukraine’s defenses.

Drones, still mostly flown by pilots, have already upended the way the war is being fought.

Ukrainian and Russian troops launch thousands of unmanned aerial vehicles (UAVs) a day at each other. Kyiv is also trying to solve its frontline troop shortage with ground robots.

The ability of drones to constantly surveil the battlefield and hit targets with precision has accelerated the “kill chain” – the process of planning and executing a strike on the enemy. AI decision-making would speed this up even more, Tsvok said.

Technological arms race

Ukraine, whose military numbers around a million personnel, is already using AI tools in its command systems.

But Tsvok said the goal was to create a single operating system to recommend battlefield decisions all the way up from individual frontline units to strategic command.

This would significantly speed up the analysis of data from the 1,200-kilometer (750-mile) front line to allow recommendations to human commanders, he said.

The aim, Tsvok said, is to unite weapons and data systems into “one single living organism that can operate in a coordinated manner.”

The technological arms race launched by Europe’s largest war since World War Two, which lasted from 1939 to 1945, has attracted interest from foreign AI companies hungry for combat data to train their models and the opportunity to test their systems.

Some, such as U.S. company Palantir, have provided Ukraine with their systems. Kyiv has created Brave1 Dataroom, a project to share battlefield data with allied countries for training their software.

“This is the place where you can understand whether your system works,” said Tsvok, wearing a black T-shirt and jeans.

Moscow is also developing its artificial intelligence capabilities. A senior Ukrainian air defense commander told Reuters in April he was concerned by Russia’s increasing use of AI in planning drone and missile attacks on cities, which could significantly reduce the planning time for each strike.

“The question is,” Tsvok said, “how quickly we build our solutions and how practically we apply them and achieve the primary impact on the battlefield from our side.”

He added that the defense ministry was developing an AI-driven recruitment and HR system as part of Fedorov’s push for data-driven reform of the vast government department.

Ukraine operates on the principle of having a human in the loop on combat decisions, but Tsvok said AI systems could eventually outrun humans, whose presence would then slow decisions down.

“Then the question arises: how do we keep up with making decisions that autonomous systems propose?” he said.

On the heels of President Donald Trump’s recent executive order on AI innovation, Emil Michael, the undersecretary of defense for research and engineering, said that he’s concerned about the category of “cyber weapons” that companies are releasing, such as Anthropic’s Mythos.

“These companies have a responsibility to ensure that their weapons, what they call weaponization potential of these models, to be careful and thoughtful about what they’re doing,” Michael said Thursday at The Washington Post’s inaugural Building America Summit.

Lawmakers have increasingly warned against the weaponization of AI models by U.S. adversaries against citizens, businesses and government agencies, calling for their crafting to include resilient security measures to safeguard against hacks.

On Tuesday, the White House released an executive order that establishes an “AI cybersecurity clearinghouse,” through which AI industry partners can volunteer to have the Defense Department scan their systems for software vulnerabilities before their release.

Michael said that companies that have models with a “weaponization capability” could allow the federal government to spend 30 days examining their systems. The government could potentially identify vulnerabilities across the country in systems with IP that could be susceptible to hacks, such as electricity grids or public hospitals.

“I think they’ve all agreed and think it’s a good idea to do that. That’s been a good constructive process,” Michael said. “I give all the companies, Open AI, even Anthropic, and Google credit for sort of agreeing that was a smart thing to do.”

Anthropic has been left out of deals with the Pentagon after the firm refused to allow unrestricted access to its Claude models for fully autonomous weapons and mass domestic surveillance.

The company sued the Trump administration over the federal government labeling the firm as a supply chain risk over its decision to restrict the military’s use of its technology.

Mythos, Anthropic’s new model, has drawn criticism as skeptics of the program point out that it could pose a danger with its hacking and cybersecurity capabilities. Anthropic previewed the model and announced that it was capable of finding ways to exploit vulnerabilities in software.

Meanwhile, the DoD has integrated AI throughout the department.

When posed a question at the summit about the government’s usage of AI, Michael said that six months ago, only about 80,000 federal employees were AI users each month. But now, there are 1.5 million, he stated, saying that the government has “raced” to ramp up usage among workers for efficiency, intelligence and warfighting.

“I think by the end of this year, I’d be shocked if three quarters of the department isn’t using AI in some way,” Michael said. “We’ve integrated all the biggest AI companies over the last few months, so, we’re, in one year, going to make progress more than the five years before it.”

Speaking at a hearing before the Senate Armed Services Committee Tuesday, Lt. Gen. Michael Conley said “necessity had been the mother of invention” in spurring the service to apply available machine learning tools to combat operations.

“On the first day, they realized that our MQ-9 [Reaper drones] and other aircraft were at risk in a very hostile environment, and they were able to take some smart people, use artificial intelligence tools, and put humans on the loop, instead of in the process the whole time, and move Top Secret national-level intel,” Conley said, referring to his “small team” of AI and tech specialists.

The process of moving intelligence would have taken human operators “20 to 30 minutes to get that to a crew, into a cockpit, or into a ground control station,” he said.

Instead, Conley said, nine crew members were able to use AI “bots” to convert applicable data to a Secret clearance level and make it available in the cockpits of aircraft — all within two to three seconds.

“We have data that indicates that we’ve saved a lot of aircraft over the last 60 days using that tool, just to provide a better battlefield situational awareness,” Conley said.

Conley also confirmed the Air Force was using its AI-powered exploitation and dissemination tools to help intelligence analysts process huge amounts of complex data, including full-motion video.

“[In] a very heavy, human-intensive process, they’ve been able to use AI again, just smart humans getting together, figuring out solutions in order to proliferate that information across the intel community in seconds, rather than what could take hours in normal processes,” he said. “So we’re learning every day, and we’re getting better every day.”

To date, two U.S. aircraft have been downed by hostile fire over Iran: an F-15E Strike Eagle fighter jet and an A-10 Thunderbolt attack plane. Three more F-15s were lost to friendly fire over Kuwait.

Reportedly, some 24 Reaper drones have been destroyed by Iran over the Strait of Hormuz and elsewhere. An Air Force F-35A Joint Strike Fighter also sustained severe damage over Iran.

While aviation losses have been significant, Conley’s comments offer a rare window into prospective casualties that may have been averted, and how AFSOC is employing emerging tools to expand its operational picture.

The Air Force recently released contracting documents for a desired “Next-Generation Air Operations Center Weapon System” that would deliver more AI tools to planners at the geographic combatant commands, an effort for which it has not specified a procurement timeline.

Conley hailed MQ-9 operations in Epic Fury particularly noteworthy.

“Our MQ-9 enterprise proved that adaptive airmen can transform any platform and mission envelope, destroying hundreds of targets in contested operating areas,” he said. “I believe recent operations have been the MQ-9 community’s finest hour.”

He also hailed the work of AFSOC in rescuing two downed F-15 pilots in April in a daring and widely praised joint operation. The mission resulted in the loss of two MC-130 aircraft as well as four Army MH-6 Little Bird helicopters. Conley said he’d like to see the aircraft replaced quickly, although he said he expected it to “take some time.”

The service’s goal is to familiarize Marines with generative artificial intelligence platforms and large language models, such as Google’s Gemini, OpenAI’s ChatGPT, and xAI’s Grok, which are among the tools available through the Defense Department’s recently launched GenAI.mil platform.

The requirement follows broader DoD efforts to integrate AI across training, administration and operational planning while establishing standards for responsible use of the technology.

In an email, Maj. Hector Infante, the communications director for Marine Training and Education Command, said the course is intended to give Marines a “foundational understanding of artificial intelligence and its relevance to today’s operating environment.”

Infante said the 45-minute online course will introduce “key AI concepts, practical application, and responsible use considerations” through interview-style segments featuring experts discussing policy and operational applications.

He added that the course will “focus on how AI can support decision-making and mission effectiveness” and promote awareness of the tool rather than technical expertise.

According to the MARADMIN, all Marines have until Dec. 31, 2026, to complete the course. They can enroll and access the course through the Marine Corps eLearning Ecosystem, or MCeLE (formerly MarineNet). Civilians working within the Corps are “highly encouraged” to take the course as well.

Additionally, the Marines said that intermediate and advanced AI courses are currently in development and will be available sometime this fiscal year.

Last December, the DoD unveiled the GenAI.mil platform as part of the White House’s plan to expand AI infrastructure and introduce AI tools throughout the government.

About a month later, the Marines made the platform its official chatbot and began retiring its previous tool, the Non-classified Internet Protocol Router Generative Pre-training Transformer, or NIPRGPT.

Kyle ‘KMo’ Moschetto, a former Marine cyber officer turned technology consultant, explained in a video tutorial that GenAI.mil makes commercial AI tools available in a secure network.

“Think of this as a military-grade version of the AI tools that you’ve heard about from ChatGPT to Anthropic Claude to Google Gemini, but it lives inside government-level security and policy enforcement,” Moschetto said.

When describing GenAI.mil, military officials often use terms like “efficient” and “force multiplier” but they’re light on practical use.

However, Defense Secretary Pete Hegseth used layman’s terms to explain the capability’s possibilities early on.

In a memo released alongside the announcement, Hegseth wrote that it “can help you write documents, ask questions, conduct deep research, format content, and unlock new possibilities across your daily workflows.”

“It removes wasted time and focuses more of our energy into decisive results for the warfighter,” he added.

The same censorship apparatus the Party built to control its people is now corrupting the AI systems its leaders depend on. The United States, by leaning into an open marketplace of information and ideas, will gain advantage as it takes a different path.

AI is increasingly training newer, faster AI models. This typically involves scraping the internet for content and then loading it into datasets for new programs. The problem: online content used for training is increasingly generated by AI. As a result, each generation of technology drifts from reality.

AI researchers refer to this as “model collapse,” a phenomenon in which models trained on their own synthetic outputs degrade over successive generations. The only defense is a constant influx of fresh, honest, human-generated information. Without it, the system folds in on itself.

China’s Great Firewall cuts off that influx, expediting the impact of model collapse within its borders. The first generation of large language models worldwide was trained on massive datasets of publicly available human-generated text. These human-derived pieces of information built an algorithmic approximation of how people think, argue, explain, and communicate.

Now, the internet is filling with AI-generated content at a rate inconceivable five years ago. Marketing copy, product descriptions, social media captions and news summaries are increasingly produced by AI systems and published online. They’ve all become generic and detached from human signals.

Outpaced by the US, China’s military places selective bets on artificial intelligence

With each cycle of new AI, the models drift further from their human origins. The newer systems amplify the patterns AI systems favor while losing nuance and amplifying existing biases. Each successive AI generation is one step further removed from the humans these programs were originally built to serve.

In China, the Great Firewall accelerates this problem. The Chinese Ministry of Public Security built the Great Firewall of China in the late 1990s to censor the Chinese people. It is now the most sophisticated information control infrastructure in human history. The Great Firewall does not just restrict what Chinese users can see. It shapes the data used to train Chinese AI systems. By design, it strips out politically sensitive events, dissenting viewpoints, and independent reporting. Left behind is a curated record of reality aligned with the Party’s narrative. That filtered record of information becomes the raw material for LLMs.

The training data fed into LLMs in China does not contain any criticism of the government, fair reportage of controversial topics, or accurate information about Chinese history. Events such as the the Uyghur detention camps exist inside the Firewall only as the state chose to describe them.

Now add model collapse on top of this. Chinese AI companies such as Baidu, Alibaba, ByteDance and dozens of others aggressively deploy AI-generated content across their platforms. This material becomes training data for the next generation of Chinese AI models. With independent reporting locked out, model collapse expedites inside the Great Firewall with no escape valve. The practical consequences of this divergence are already visible and will accelerate.

Chinese LLMs struggle with tasks that require new observation, original synthesis or human complexity that the training data was designed to suppress. If asked about repression in Chinese history or controversial current events such as the Uyghur detentions, these LLMs either do not answer or produce a response indistinguishable from a Party press release. A Chinese trade official relying on domestic AI to model the economic impact of Western sanctions is working from a system incapable of providing an honest account of how those sanctions functioned or failed in comparable historical cases.

The West has a milder version of this problem.

Western AI models are trained on increasingly synthetic content, but human reporters regularly push new information into the ecosystem. Free and open societies have structural advantages to retain the capacity to reason about what is happening in the world rather than what previous AI systems described. When asked about Tiananmen Square, Western accounts typically focus on the 1989 protests and the crackdown. By contrast, Chinese models either refuse to answer or return state-aligned language. This information gap becomes part of the training data for the next generation.

To maintain a competitive AI advantage over China, Washington should treat human-generated data as a strategic asset and invest in journalism, open web archives, and synthetic-content labeling. Preserving the integrity of American training data is a defense imperative, not a tech problem.

Chinese leaders deploying AI products to make decisions about economics, geopolitics, and public health will make those decisions based on systems trained on what China’s information control apparatus wants people to believe. That is not an intelligence system. It is a mirror. And the tragedy of model collapse is that a mirror that has been looking at itself long enough no longer reflects anything.

Joe Buccino is a retired U.S. Army colonel and the author of “When Every Word Counts: How to Earn Trust, Command Attention, and Communicate Clearly in Any Situation.”

The move has stakeholders urging not to lean too hard on algorithms to distinguish, for instance, a scientist-spy sharing secret nano-energy plans with China from a Chinese professor publishing AI safety studies. Hanging in the balance lies troops’ technological edge, veteran intelligence officials and academics say.

AI’s confusion over the timing and nature of research partnerships may obscure real espionage if humans are not the ultimate judge of foreign influence, they warn, with some academics fearing an AI-work slop redux of the “China Initiative,” where the first Trump administration charged dozens of ethnic Chinese scientists with fraud, only to drop nearly all charges.

“Automated vetting tools are extremely useful for vetting large datasets and identifying patterns of concern. But those tools are for decision support to help the people, the human analysts, assess context and intent,” said David Cattler, who, until September, led the Defense Counterintelligence and Security Agency, which screens personnel seeking security clearances.

The renewed focus on academic research security comes as tensions between the U.S. and China over AI intensify, with China ramping up both the development of low-cost AI models and its alleged exploitation of U.S. models.

The U.S. intelligence community’s annual threat assessment reports that China “aims to displace the U.S. as the global AI leader by 2030,” in part, “by using its sizeable talent pool, extensive datasets, government funding and burgeoning global partnerships.”

Partnership is a complicated term in military “fundamental research” — studies that are publishable, rather than proprietary or classified, with potential defense applications.

For instance, mere co-authorship with a China-based scientist does not suffice to deny a U.S. researcher money, according to evolving academic research rules targeting foreign influence. Rather, to make the call, the Pentagon must assess each academic’s disclosures of external funding sources and affiliations.

According to a recently-declassified May 2025 inspector general report, such disclosures went unchecked because the Pentagon had not “requested additional government full-time equivalent employees to thoroughly review… and to conduct oversight of over 27,000 academic research awards.”

Defense News first obtained the report and the Pentagon’s response to a draft version through an open records request.

When asked about additional staffing, a Pentagon official pointed to orders in a January research security directive for the Chief Digital and AI Office (CDAO) to identify “automated vetting and continuous monitoring capabilities” and create a common research grant database.

The mandate also calls for a year-long “damage assessment” of selected research transactions, including cases that the House Select Committee on China flagged, in part, using AI tools.

A September GOP-led Committee report alleged that the Pentagon subsidized 1,400 academic papers published between June 2023 and June 2025 involving partnerships with the Chinese government.

The Pentagon said in an emailed statement on Thursday that the department “is committed to protecting the integrity of U.S. research while fostering international collaboration. Our approach leverages advanced analytical tools to augment human expertise, ensuring a rigorous and fair review process.”

Machines Make the Same Mistakes

Several university representatives, who requested anonymity due to the sensitivity of policy discussions, note that AI-assisted risk assessments have, in the past, drawn false assumptions about U.S.-China research collaborations.

For instance, the AI-aided GOP House report mislabeled the state-backed Wuhan National Laboratory for Optoelectronics as the sponsor of a collection of nine essays, whereas the Wuhan lab supported only one essay that involved no U.S. authors or federal funding.

The report also mistook a Chinese military-affiliated author’s publication on single-electron transistors for a DOD-funded project, when it merely referenced work funded by the U.S. military.

House Select Committee officials declined to comment on the cases.

Averting foreign interference “is not as easy as just having better AI capacities, because we know that some of those AI mistakes were the same human mistakes that led to inaccurate charges brought against researchers under the China Initiative,” Toby Smith, senior vice president for government relations and public policy at the Association of American Universities, said after learning of the AI gaffes.

About 30% of the China Initiative’s 77 cases involved academics not disclosing Chinese partnerships or funding sources, though disclosure was often not required, or no such partners or funding existed.

A jury found only one professor, Harvard nanochemist Charles Lieber, guilty, after he lied to Pentagon investigators about his participation in a Chinese talent recruitment program and made other false statements.

Smith said that relying on proxies — such as co-authorship, affiliation or nationality — to deduce security concerns led investigators to confuse co-authorship with direct interaction or shared data access, and to misinterpret historical ties as active partnerships.

“AI systems trained on bibliometric data,” or citation analytics, “and affiliation records can inherit the same flawed assumptions that underpinned the China Initiative,” he said. “The core lesson from the China Initiative is that identifying genuine research security risk requires judgment, context and proportionality — qualities that automated systems should support, not replace.”

Gisela Perez Kusakawa, executive director of the Asian-American Scholar Forum, said that the Pentagon has not communicated the types of data that the new AI tools will collect, use and share, nor sought community feedback, heightening concerns about ethnic profiling.

In an emailed statement, the Pentagon responded, “It is the standard practice of the department to not provide specifics regarding the criteria and weighting for threat assessments conducted, whether by manual or automated process.”

Kusakawa said, “We should be making sure that our research environment is welcoming, that we are encouraging these talents, especially in AI, to come to the United States and, frankly, have family in the United States, and make this a country that they contribute to and invest their and their children’s future in.”

Increasingly, that AI talent is not coming. Statistics suggest that Chinese AI experts are staying in China, even as America has managed to keep earlier expats.

Of about 100 Chinese-origin AI scholars who were researching at U.S. institutions in 2019 — when their papers were accepted at the world’s most elite AI conference, NeurIPS — 87% remained stateside as of 2025, according to Carnegie Endowment for International Peace figures.

At the same time, the share of Chinese home-grown talent not moving abroad has skyrocketed since 2019, based on an Economist tally of similar data. In 2019, about 30% of NeurIPS authors educated in China were still in China. By 2022, that population had jumped to 58%, and up to 68% in 2025.

‘Key’ Financial Disclosures Were Missing

To ensure that no country has a stranglehold on general knowledge, “the goal cannot be to close the U.S. research system altogether,” said Cattler, the former Defense Counterintelligence and Security Agency director and now founder of consultancy Ironhelm Works. “The goal must be to ensure that collaboration strengthens national security rather than inadvertently weakening it.”

Last year’s inspector general report concluded that Pentagon agencies that partner with academic institutions “could be at an increased risk of exposure to foreign influence” because military units were “missing key documents” that can help discern scientists’ foreign financial sources, outside employers and other details related to potential conflicts.

For instance, about 80% of Air Force funding transactions sampled in the report were missing documents that can reveal problematic relationships. Also, the Air Force and the Defense Advanced Research Projects Agency — which bankrolled the Internet’s inventors — told evaluators that they had not, as required, annually checked that researchers named in progress reports were not involved in banned talent recruitment programs.

In response to questions about the screening lapses, the Pentagon said in an emailed statement that the department “is aware of the OIG’s findings in the report and the impacted directorates are working with [the Office of the Undersecretary for Defense and Engineering] to address the items identified.”

Jeffrey Stoff, a Chinese linguist and intelligence community analyst for 18 years who resigned in 2021 due to frustration with research security, maintains that DOD still needs more human expertise in language, culture and the minutiae of research restrictions.

“AI can and should be used for unsophisticated, labor-intensive tasks,” such as cross-referencing foreign organizations in financial disclosure forms against the names of blacklisted parties, but humans still need to inspect AI’s work, as not all nefarious ties are machine-readable, said Stoff, who now advocates for tighter safeguards as head of the nonprofit Center for Research Security and Integrity.

Cattler, giving more credit to AI, said that the scale of research collaboration demands that the Pentagon upgrade its approach to clocking potential spies.

With such a large population of research institutions and affiliated scientists, as well as a bombardment of often duplicative alerts, automation “improves the signal within that noise and can help orient the humans on really important matters,” he said.

“Sometimes people are deliberately deceptive when they are processed, and that could happen in a human exchange just as much as it could happen in something that a computer can see, but together, a human review and a computer review are incredibly powerful.”

The armed forces are working with partners on a system to provide what Desmeulles called “true distributed working capability” centered on data and using advanced AI, “a sovereign system that will essentially be the equivalent of Maven.”

The system could be available within a few months, and available for exercises in September 2027, the general said, declining to provide specifics.

Project Maven is a Pentagon program that uses AI to process drone and surveillance data to automatically detect and track objects, using technology provided by contractors including Palantir Technologies. Maven has faced controversy amid questions about AI-assisted targeting in Iran, with concerns about speed, accountability, and harm to civilians related to automated kill chains.

“We’ve really positioned data as the center of everything we do,” Desmeulles said in a briefing with three reporters on Saturday at the Montmorillon military camp in western France, describing data as the ammunition of the command post.

“The centrality of data is something that’s well understood by the corps, the Army, and the French forces,” he said. “So, we’re really focused on that.”

The armed forces are on track to develop “a true distributed working capability, based on highly advanced artificial intelligence and centered on data,” Desmeulles said. “We’re following that logic, to remain sovereign, and that’s an area where we are strong.”

Desmeulles said his corps is already seeing “very, very good” results from a data-centric approach, even if there is “still a little way to go before it’s practically perfect in my eyes.”

France has several AI companies that are active in defense, including Comand AI, ChapsVision and Safran’s AI business, and is also home to a major developer of large-language models with Mistral AI. France in 2024 created an agency under the Armed Forces Ministry that works on AI for defense.

The website cited a state-run media report and experts calling the vessel a “major leap in integrated combat capability” that “positions the vessel among the most advanced frigates in service today”.

A slew of announcements such as that one from March 30 shows AI expanding across a military that aims to “intelligentize” as it prepares for potential conflicts in the South China Sea and Taiwan Strait. But analysts say China is picking its AI battles carefully rather than expecting quick domination of the technology or short-term parity with the United States.

China is taking a “cautious official posture” toward AI in the armed forces, said Sophie Wushuang Yi, postdoctoral teaching fellow with Schwarzman College at Tsinghua University.

“China’s concept of intelligentized warfare has been embedded in official defense white papers since 2019,” Yi said. “But the open-source academic literature is frank that China cannot currently close the overall gap with the United States in military AI capability.”

Still, AI is becoming a force within the forces.

An institution under the People’s Liberation Army in January used AI to test drone swarms and, according to a test run shown on Chinese state television, one soldier supervised some 200 of the autonomous vehicles at the same time.

AI is taking on a greater role as well in the military’s use of space and cyberspace, said Malcolm Davis, a senior analyst for defense strategy and national security with the Australian Strategic Policy Institute. In space, he said, it can manage “complex orbital operations,” while in cyberspace it can plan and conduct operations against critical information infrastructure.

The military’s ability to use AI at machine speed would potentially let it exploit a faster “observe-orient-decide-act” loop compared to purely human-controlled systems, Davis said.

“That’s something that’s being demonstrated by the U.S. and Israel now in operational planning in the Iran war, where AI is playing a key role in identifying targets and planning mission packages,” the Canberra-based analyst said. “There’s no reason that the PLA won’t learn from that and utilize a similar capability.”

A testament to AI’s reach throughout the military, a March 26 PLA Daily report notes its use in battlefield perception, intelligent decision support and autonomous control systems.

PLA leaders particularly value AI decision-making because most of their people lack battlefield experience, unlike American counterparts, said Sam Bresnick, a research fellow with the Center for Security and Emerging Technology at Georgetown University.

He said its priorities include layering AI on top of computer networks, gathering volumes of data and the autonomy of unmanned systems such as uncrewed underwater vehicles.

Chinese officials want to surpass the U.S. in military AI use, Bresnick noted, but the government today fears information that AI could use or generate. “The data could go against Xi Jinping and Communist Party ideals,” he said. “They don’t want to lose control over it.”

The U.S. armed forces now have a “commanding” AI lead over China, the Modern War Institute at West Point said in a March 17 study.

It says the United States has more than 4,000 data centers versus some 400 in China. Four-year-old U.S. export controls on advanced semiconductors shipped to China limit Beijing’s access to AI-related hardware, the study adds.

“China’s publicly stated position is considerably more cautious and more hedged than is commonly assumed in Western coverage,” Yi said.

“The PLA lacks the volume of real operational data that the U.S. military has accumulated over decades of expeditionary warfare, and there are unresolved doctrinal tensions between the decentralized decision-making that effective AI-enabled operations require and the PLA’s deeply embedded centralized command culture,” said Yi of the Schwarzman College.

China may have surpassed the United States in AI for drone swarms, however, said Chen Yi-fan, assistant professor in the Diplomacy and International Relations Department at Tamkang University in Taiwan.

“With the addition of drone carriers already in service, the PLA has taken the lead over the U.S. military in this category of AI military applications,” he said.

The Qinzhou frigate was commissioned last year and did a combat drill in the South China Sea, where Beijing disputes maritime sovereignty with five other governments.

Lt. Gen. Christian Freuding, who became army chief last October after years of overseeing Germany’s arms supplies to Kyiv, shared his insights from visits to Ukrainian command posts where drones and modern sensors have drastically increased the volume of battlefield data.

“The Ukrainians exploit data which they have collected over four years of war. Based on this data, the AI can deduce how the enemy has acted in similar situations in the past - and recommend countermeasures,” he said.

He noted that tasks now requiring hundreds of personnel and days to complete could be sped up significantly through AI, adding that conventional methods alone would never be enough to “break the adversary’s decision-making cycle.”

Freuding suggested utilizing data from Ukraine and from German military exercises when training analytical tools, ensuring alignment with Germany’s operational principles.

Addressing ethical concerns, he emphasized that AI would serve only as an advisory tool to facilitate human decision-making.

“The task of taking analytical and balanced decisions will always remain with the human, with the soldier,” he said, adding that while a specific AI product is yet to be selected, the technology’s deployment is a priority.

Freuding underscored the importance of aligning Germany’s AI systems with NATO’s evolving standards. He did not rule out a European-developed system, but said American solutions might offer practical advantages due to their advanced deployment.

“Personally, I think it’s important that we get something up and running quickly. Of course, issues like data sovereignty and security need to be taken into account,” he added.

The U.S. Army is fielding the AI tool Maven, made by the Silicon Valley company Palantir, to process battlefield data, including imagery and video, to improve situational awareness and speed up decision-making.

Defense officials in the Trump administration also warned they could designate Anthropic, which makes the AI chatbot Claude, as a supply chain risk — or invoke a Cold War-era law called the Defense Production Act to give the military more sweeping authority to use its products, even if the company doesn’t approve.

Some experts say that using the law this way would be unprecedented, and could bring future legal challenges. The government’s efforts to essentially force Anthropic’s hand also underscore a wider, contentious debate over AI’s role in national security.

Here’s what we know.

What is the Defense Production Act?

The Defense Production Act gives the federal government broad authority to direct private companies to meet the needs of national defense.

The act was signed by President Harry S. Truman in 1950, amid concerns about supplies and equipment during the Korean War. But over its now decades-long history, the law’s powers have been invoked not only in times of war but also for domestic emergency preparedness as well as recovery from terrorist attacks and natural disasters.

One of the act’s provisions allows the president to require companies to prioritize government contracts and orders deemed necessary for national defense, with the goal of ensuring the private sector is producing enough goods needed to meet a war effort or other national emergency.

Other provisions give the president the ability to use loans and additional incentives to increase production of critical goods, and authorize the government to establish voluntary agreements with private industry.

The DPA is “one of the government’s most powerful and adaptable industrial policy tools,” said Joel Dodge, an attorney and the director of industrial policy and economic security at the Vanderbilt Policy Accelerator.

Anthropic is the last of its AI peers to not supply its technology to a new U.S. military internal network. Its CEO Dario Amodei repeatedly has made clear his ethical concerns about unchecked government use of AI, including the dangers of fully autonomous armed drones and of AI-assisted mass surveillance that could track dissent.

The Defense Department is considering invoking the DPA to give the military more authority to use Anthropic’s products, even if the company doesn’t approve of how, according to a person familiar with the matter and a senior Pentagon official.

That could mean forcing Anthropic to adapt its model to the Pentagon’s needs without built-in safety limits, or remove certain ethical restrictions from the company’s contract language.

Experts like Dodge say both would be “without precedent under the history of the DPA.”

“It’s a powerful law,” he said. ”(But) it has never been used to compel a company to produce a product that it’s deemed unsafe, or to dictate its terms of service.”

How has this law been used in the past?

Trump in his first term and former President Joe Biden invoked the DPA to boost supplies to combat the COVID-19 pandemic. And during 2022’s nationwide baby formula shortage, Biden used the law to speed production of formula and authorize flights to import supply from overseas.

Biden also invoked the DPA in a 2023 executive order on AI, notably in efforts to require that companies share safety test results and other information with the government. Trump repealed the order at the start of his second term.

Decades ago, the administrations of both President Bill Clinton and George W. Bush used the DPA to ensure that electricity and natural gas shippers continued supplying California utilities amid an energy crisis.

And the law was used after Hurricane Maria struck Puerto Rico in 2017 to prioritize contracts for food, bottled water, manufactured housing units and the restoration of electrical systems.

The DPA requires periodic reauthorization to remain in effect, which can expand or refine the scope of the law. According to congressional documents, its next expiration date is slated for Sept. 30 of this year. And depending on how the Defense Department’s reported demands unfold, Anthropic could be at the top of lawmakers’ minds.

Possible next steps for Anthropic

If the Defense Department uses the DPA provision aimed at prioritizing government contracts and ordering production of certain goods — which the Anthropic case suggests it will — a company can push back if the requested product isn’t something it already produces, Dodge and others say, or if it deems the terms to be unreasonable.

But the government may try and overrule that, notes Charlie Bullock, senior research fellow at the Institute for Law & AI.

“If neither side backs down, it seems realistic that there would be litigation between Anthropic and the government,” Bullock said.

Some have also noted tension between the Pentagon’s warning that it could designate Anthropic as a supply chain risk while also indicating that its products are so important to national defense that it needs to invoke the DPA — two assertions that seem at odds with each other.

“There are a lot of forces that I think the administration’s counting on that would lead Anthropic to just give in on Friday and agree with its terms,” Dodge said.

If there’s future litigation over a potential DPA order, Dodge doesn’t expect the government to prevail because “it seems very out of bounds under the text of the law.”

But if the administration is successful, or Anthropic simply agrees to new terms, that could open up “a Pandora’s box of what the government could do to assert power and control over private companies,” he added.

The goal of the new GenWar lab, scheduled to open in 2026 at the Johns Hopkins Applied Physics Laboratory in Laurel, Maryland, is to improve tabletop exercises by harnessing the speed and user-friendliness of large language models, or LLMs, which include popular chatbots such as ChatGPT.

By accessing AI during an exercise, human players can quickly experiment with different strategies. At the same time, human participants will be assisted by AI agents playing the role of staff advisers or even enemy leaders.

But the GenWar lab will offer even more intriguing possibilities. One is to allow human players to interact directly with sophisticated computer models operating behind the scenes of tabletop exercises. Another is the possibility of wargames played solely by AI actors on both sides.

“We’ve heard a demand signal from our sponsors of the need to do wargaming faster,” Kevin Mather, who heads the GenWar Lab, told Defense News. “The ability to get more in depth, the ability sometimes to be able to include modeling and simulation and do what-if analysis.”

Since the Prussian Army began using “kriegsspiel” to train staff officers in the 19th Century, tabletop exercises have essentially pitted Blue against Red teams, with an umpire to judge.

The problem is that these wargames are labor-intensive to design and adjudicate, and too cumbersome to allow replay and incorporating lessons learned.

But AI could yield multiple iterations of a game, or allow a scenario to be redone. For example, players might choose a strategy, only to have a subject matter expert rule that the idea wasn’t realistic.

“Let’s rewind the gameplay and go back one turn,” said GenWar Lab program manager Kelly Diaz, who discussed a hypothetical scenario with Defense News. “We’re going to retry that move. And because it’s all digital, we’ll have a log. Afterward for the post-game analytics, we can kind of trace through how the decisions were being made.”

To facilitate this, GenWar Lab uses an array of tools. GenWar TTX creates the digital environment and AI agents for the exercise. GenWar Sim — built on the government-owned Advanced Framework for Simulation, Integration and Modeling, or AFSIM — allows players to interact with physics-based models used for adjudication.

Thus, GenWar Sim functions as a translator that lets humans communicate in plain speech with the mathematical models operating behind the scenes.

“As a human, you give your commands: ‘I’d like to attack here, I’d like to defend there,’” explained Mather. “We’ve built code in the modeling and sim engine to read from that database layer and automatically execute those commands.”

Conversely, the LLMs can communicate with humans in ordinary speech.

Still, the thought of AI players may give some humans pause. As anyone who has played commercial strategy games can attest, computer players are not the sharpest of opponents.

“We won’t remotely claim that they’re making optimal decisions,” Mather said.

However, the hope is that their decisions are realistic enough to facilitate gameplay and allow the players to explore multiple strategies during an exercise.

Mather and Kelly emphasize that AI will not replace traditional wargaming techniques.

“It’s not nearly as in depth, for example, as a traditional ops analysis or modeling and sim study,” Mather said. But AI can offer “those 70% to 80% solutions that are not the answer, but really accelerate the human learning,” he added.

As AI is woven into societal fabric, general consensus is that it will inevitably become part of wargaming. The question is whether AI will dominate in the space.

Benjamin Jensen, a researcher at the Center for International and Strategic, believes AI can enhance wargames — if properly documented and evaluated.

The risk is that “we find strategic analysis reduced to, ‘Here is what an LLM said,’” he told Defense News.

How well LLMs relate to national security policy is also unclear.

“The larger challenge is that most foundation models commonly used haven’t been sufficiently benchmarked against strategy and statecraft,” Jensen said. “So, using AI to support game design, development and execution is a great idea. The question is how far that use goes and how well it is documented to avoid common pitfalls.”

Following previous DIU Director Doug Beck’s sudden resignation from the role Monday, the Pentagon announced Undersecretary of Defense for Research and Engineering Emil Michael would lead the organization in an acting capacity. While the decision at least temporarily puts DIU under Michael’s purview, he was adamant the move was not permanent.

“DIU is a separate organization independently created in law with its own mission set,” Michael told reporters on the sidelines of a National Defense Industrial Association conference in Washington, D.C. “I’m acting until we find a full-time person.”

When DIU was first established in 2015 as a Defense Department outpost for Silicon Valley tech firms, it reported to the Pentagon’s head of research and engineering — a construct that senior leaders said impeded its influence. The following year, then-Defense Secretary Ash Carter elevated the office to report directly to him.

By 2018, it was shuffled back within the research and engineering portfolio, where it remained until 2023. At that time, former Secretary of Defense Lloyd Austin again gave it a direct reporting line. Congress codified the office’s structure in the fiscal 2024 National Defense Authorization Act, which states that DIU’s director “shall report directly to the Secretary without intervening authority.”

While Michael won’t permanently wear the DIU director hat, at least for now, the Pentagon did recently expand his office’s scope. The department announced Aug. 20 that the Chief Digital and AI Office, formerly an independent organization, would reside within research and engineering moving forward.

At the time, the Pentagon said the move would “accelerate Department-wide AI transformation.” Michael reiterated that message Wednesday.

“Having it be part of R&E means that it’s part of a larger organization that hopefully has the muscle and the wherewithal to further develop and do bigger partnerships ... and then more easily proliferate that technologies into various areas of the department,” he said.

Pentagon CTO to oversee Defense Innovation Unit following Beck’s exit

Asked how his organization might spearhead AI advances that could support Golden Dome — a DOD initiative to create an advanced, multi-layered homeland missile defense shield — Michael was careful to note he doesn’t have direct involvement in the project. However, he said his intent is to invest in AI platforms that could be made available for large-scale efforts like Golden Dome, as well as for combatant commands, Pentagon employees and even defense contractors to support their mission.

“It’s really [about] trying to use the latent capability that’s primarily been invested in by the private industry,” he said. “They’ve been putting $300 billion per year into AI, whether it be software development, data centers and so on. So how does the DOD leverage that for an endless list of use cases that are beneficial.”

Along his focus on harnessing AI for defense applications, Michael noted that his team is also crafting a list of critical technology focus areas — a practice that has become a tradition for incoming research and engineering chiefs. He noted that rather than add to current list, which includes some 14 technologies, he’s working to narrow it to include only the department’s highest priorities.

Rather than craft roadmaps for each area, Michael added, he’ll instead call for “sprints” that aim to accelerate broader technology development and adoption.

“It’s a pyramid hierarchy, if you will. These are the most important things I need people to wake up every day thinking about,” Michael said. “I want to communicate that to industry.”

Each company received a contract worth up to $200 million, according to a notice from the Chief Digital and AI Office. The firms will help the Defense Department develop agentic AI workflows for key national security missions.

“Leveraging commercially available solutions into an integrated capabilities approach will accelerate the use of advanced AI as part of our Joint mission essential tasks in our warfighting domain as well as intelligence, business, and enterprise information systems,” Chief Digital and AI Officer Doug Matty said in a statement.

The military services have adopted generative AI tools to varying degrees and for a range of tasks — from tech support to finding files. Agentic AI uses more advanced reasoning to address and act on more complex challenges.

The Pentagon didn’t specify what missions the program would support, but the department has said it wants to use AI in areas like intelligence analysis, campaigning, logistics and data collection.

Following the announcement, Elon Musk-owned xAI — whose conversational AI chatbot goes by the name Grok — unveiled a U.S. government-specific production line called Grok for Government.

Announcing Grok for Government - a suite of products that make our frontier models available to United States Government customers

We are especially excited about two new partnerships for our US Government partners

1) a new contract from the US Department of Defense
2) our…

— xAI (@xai) July 14, 2025

Grok has come under scrutiny after an update generated a slew of racists and antisemitic comments. In one instance, the chatbot referred to itself as “MechaHitler.”

The award follows Musk’s months-long push from within the White House to slash federal spending. Amid a public falling out earlier this summer, President Donald Trump has threatened to cancel government contracts awarded to Musk’s companies. Under this new deal, however, the billionaire’s federal work would expand.

The company, Obviant, was founded in 2023 and developed its defense acquisition platform as a means to help both DOD and the defense industry better understand how the Pentagon is spending its money and what stakeholders in Congress and in the department are prioritizing.

Navigating that system typically means combing through the thousands of pages of PDFs that make up the Defense Department’s budget request, tracking congressional markups through multiple committees and — in the fiscal 2026 cycle — keeping tabs on the status of a $150 billion budget reconciliation bill.

“There’s no source of truth. There’s no mapping for that,” Obviant founder and CEO Brendan Karp told Defense News in an interview. “All this information is just loads of structured and unstructured sources.”

On Tuesday, Obviant announced $7 million in seed funding that it will use to increase the AI-based platform’s capabilities and make it available to more users.

“With this recent funding, Obviant will expand its team, advance the AI capabilities of its platform, and broaden its customer base — transforming how companies, investors, and government leaders make more informed decisions across the defense sector,” Obviant said in a statement.

The funding round was led by Shield Capital, a venture firm led by former DOD and defense industry experts focused on early-stage companies developing AI, autonomy, cyber and space capabilities.

Karp and his team rolled out their platform last year and have targeted a few main customer groups: investors trying to understand where the department is focusing its funding, companies trying to sell to the military, congressional staff tracking DOD funding priorities and program offices and acquisition leaders in the Pentagon trying to transition technology to the field.

For all of these users, Karp said, Obviant pulls data from open-source budget documents and appropriations and contract announcements, as well as any unique customer data, and finds common threads within it.

The company provides information about how much the department is investing in certain capabilities or using certain contracting tools. It can also track trends, such as whether DOD’s public rhetoric matches its actual spending.

The firm has experienced success in its first few years, recently closing its first government contracts with several DOD innovation organizations, which Karp said he can’t yet disclose.

As lawmakers call for reforms to the defense acquisition system and Defense Secretary Pete Hegseth pushes for greater speed and efficiency in the weapons-buying process, Karp said he sees a role for the company in facilitating progress in those areas.

“We feel like we can play an accelerant role in that process,” he said.

The situation arrives as concerns grow that lax regulations are allowing AI programmers to dodge responsibility for an algorithmic actor’s perpetration of emotional abuse or “no-marks” cybertorture. Notably, a teenager allegedly died by suicide — and several others endured mental distress — after conversing with self-learning voicebot and chatbot “companions” that dispensed antagonizing language.

Now, seven years after writing about physical torture in “The Rise of A.I. Interrogation in the Dawn of Autonomous Robots and the Need for an Additional Protocol to the U.N. Convention Against Torture,” privacy attorney Amanda McAllister Novak sees an even greater need for bans and criminal repercussions.

In the wake of the use of “chatbots, voicebots, and other AI-enabled technologies for psychologically-abusive purposes — we have to think about if it’s worth having AI-questioning systems in the military context,” said Novak, who also serves on the World Without Genocide board of directors. Novak spoke to Defense News on her own behalf.

Royal Reff, a spokesperson for the U.S. Defense Counterintelligence and Security Agency (DCSA), which screens individuals holding or applying for security clearances, said in a statement that the agency’s AI voice agent was “expressly developed” to “mitigat[e] gender and cultural biases” that human interviewers may harbor.

Also, “because systems like the interviewer software” are “currently being used worldwide, including within the U.S. by state and local law enforcement, we believe that it is important for us to understand the capabilities of such systems,” he stated.

DCSA spokesperson Cindy McGovern added that the development of this technology continues because of its “potential for enhancing U.S. national security, regardless of other foreign applications.”

After hearing the rationale for the project, Novak registered concerns about the global race for dominance in military AI capabilities.

Odds are high that “no matter how much the government cares about proper training, oversight and guardrails,” cybercriminal organizations or foreign-sponsored cybercriminals may hack into and weaponize military-owned AI to psychologically hurt officers or civilians, she said.

“Online criminal spaces are thriving” and “harming the most vulnerable people,” Novak added.

Investors are betting $500 billion that data centers for running AI applications will ultimately secure world leadership in AI and cost savings across the public and private sectors. The $13 billion conversational AI market alone will nearly quadruple to $50 billion by 2030, as the voice generator industry soars from $3 billion to an expected $40 billion by 2032. Meanwhile, the U.S. Central Intelligence Agency has been toying with AI interrogators since at least the early 1980s.

DCSA officials publicly wrote in 2022 that whether a security interview can be fully automated remains an open question. Preliminary results from mock questioning sessions “are encouraging,” officials noted, underscoring benefits such as “longer, more naturalistic types of interview formats.”

Reff stated that the agency “would not employ any such system that was not first sufficiently studied to understand its potential,” including “negative consequences.”

Researchers let volunteers raise concerns after each test, and none have indicated that the voicebot interviewer harmed them psychologically or otherwise, he added.

However, experiments with a recent prototype barred volunteers with mental disorders from participating, whereas actual security interviews typically do not exempt individuals with mental health conditions from questioning, noted Renée Cummings, a data science professor and criminal psychologist at the University of Virginia.

Cummings said that research and testing can never account for every psychological variable that may influence the emotional or cognitive state of the interviewee.

In laboratory or field-testing scenarios, “you can’t expose someone to extreme torture by an avatar or by a bot to see the result,” said Cummings, who also sits on the advisory council of the AI & Equality Initiative at the Carnegie Council for Ethics in International Affairs. “We need to focus on gaining a more sophisticated understanding about the impacts of algorithms on the psychological state, our emotions and our brain — because of AI’s ability to impact in real-time.”

Screening the voicebot screeners

An early 1980s CIA transcript of a purported conversation between an AI program and an alleged spy highlights the national security world’s long-held interest in automating human-source intelligence, or HUMINT, collection.

Developers designed the chatbot’s algorithms to analyze correlations between suspected spook “Joe Hardesty’s” behaviors and words or phrases in Hardesty’s responses to open-ended questions.

From this analysis, the chatbot pegged some of the suspect’s vulnerabilities, including topics that may hit a raw nerve:

AI: What do you tend to think is my interest in your government?

Hardesty: I think you would do in my government if you had a chance.

AI: You are prone to think I would do in your government because?

Hardesty: You are anti-democratic.

In 1983, the CIA quipped that Joe Hardesty is fortunate that “should the probing get too discomforting, he will have an option that will not be available to him in a true overseas interview situation — he can stop the questions with a flick of the ‘off’ switch.”

Today, Bradley Moss, a lawyer who represents whistleblowers accused of leaking, is not laughing at DCSA’s vision of fully-automated interviews.

“There would have to be someone observing” the avatar “that can stop the process, that can come into the room and insert a human element to it,” he cautioned.

Then again, other critics note that a human may not want to intervene. A situation with “a human overseer monitoring the chatbot is fine, unless the human also believes that verbal torture is a good thing. And that is a real possibility,” said Herb Lin, a senior research scholar for cyber policy and security at Stanford University who once held several security clearances.

Despite a global ban on torture and evidence that rapport-building interviews are more efficient than ones gunning for confessions, intelligence officials continue subjecting interviewees to inhumane treatment.

Novak, the data privacy attorney and anti-genocide activist, said the idea that excessive force gets the job done may encourage some government contractors to set performance metrics for voice algorithms that permit bullying.

“While AI has a tremendous ability to bring out the best in human potential, it also has the ability to exacerbate the worst of human potential and to elevate that hostility,” she added.

Parents contend that voicebot and chatbot companions, trained on content from online forums, show just that.

Character AI, the firm that offers virtual relationships with bots patterned after icons and influencers, profits from users glued to their phones. Each user, on average, talks two hours a day with various virtual friends. Character AI officials have stated that, after learning of concerns, the company reduced the likelihood of users encountering suggestive content and now warns them of hour-long conversations.

Novak commented that the accusations against Character AI illustrate the ability of self-learning voicebots to amplify indecency in training data and to distance trainers from accountability for that offensiveness.

In the national security realm, developers of military AI voice systems that may potentially abuse or terrorize operate in a virtually lawless no-man’s land. AI ethics policies in the world’s most powerful country offer nearly no guidance on de-escalating abusive conversations.

Early U.S. responsible and ethical AI policies in the national security space focused on accuracy, privacy and minimizing racial and ethnic biases — without regard for psychological impact. That policy gap widened after President Donald Trump revoked many regulations in an executive order aimed at accelerating private sector innovation.

In general, psychological suffering is often “brushed away” as a mere allegation since “it leaves no obvious physical scars.”

The United Nations initially did not address “psychological torture” under the 1984 Convention Against Torture. Not until five years ago did the term take on significance.

A 2020 UN report now defines “torture,” in part, as any technique intended or designed to purposefully inflict severe mental pain or suffering “without using the conduit” of “severe physical pain or suffering.”

At bottom, psychological torture targets basic human needs. The interrogation techniques arouse fear; breach privacy or sexual integrity to kindle shame or suicidal ideation; alter sound to heighten disorientation and exhaustion; foster and then betray sympathy to isolate; and instill helplessness.

When an algorithmic audiobot is the conduit of cruelty, neither the UN treaty nor other international laws are equipped to hold humans accountable, Novak said. Without binding rules on the use of AI in the military context, “bad actors will deny any such torture was ‘intentional’” — a required element for criminal responsibility that the law created before AI matured, she explained.

Looking to the future, Nils Melzer, author of the 2020 UN report and a former UN special rapporteur on torture, warned that the interpretation of psychological torture must evolve in sync with emerging technologies, “such as artificial intelligence,” because cyber environments provide “virtually guaranteed anonymity and almost complete impunity” to offenders.