China’s Military Says AI Can’t Replace Commanders. Xi Is Testing That

For years, the Chinese military’s own newspaper has been making a promise it may no longer be able to keep. Whenever the People’s Liberation Army Daily (Jiefangjun Bao), the official newspaper of China’s Central Military Commission, discusses artificial intelligence and command, it returns to the same reassurance: Machines may sort sensor data, draft options, and compress the time between observation and action, but the final authority, and the responsibility that comes with it, belong to the human commander. As one 2025 article puts it, AI cannot resolve the allocation of command authority and responsibility because only a human commander can bear the risk and the responsibility of war.On Jan. 24, 2026, China’s Ministry of National Defense announced that Zhang Youxia, vice chairman of the Central Military Commission, and Liu Zhenli, its chief of the Joint Staff Department, were under investigation for serious violations of discipline and law. They were the last two men at the apex of the Chinese military with substantial operational experience. With their removal, five of the six members General Secretary Xi Jinping appointed to the commission in 2022 are gone. Two remain: Xi, a military amateur, and Zhang Shengmin, a career discipline-inspection officer who has never commanded troops in the field. The chief of the Joint Staff Department’s chair, the nerve center of operational planning, has been left vacant for the first time since the Cultural Revolution.The People’s Liberation Army is insisting on the irreplaceability of experienced commanders at precisely the moment it is running out of them. That contradiction is the subject of this essay. The purges and the military’s accelerating embrace of artificial intelligence are usually treated as separate stories, but they are two expressions of a single problem: how Xi preserves control over an army whose operational competence he does not fully trust. The purges reflect distrust of human commanders. The turn to AI expresses hopes in the machines meant to compensate for them. Together, they suggest the hollowing out of the Central Military Commission is not simply a crisis Xi has failed to resolve, but one he appears willing to tolerate.A Promise Made at ScaleThis is a message the People’s Liberation Army Daily repeats systematically and with increasing frequency. In a corpus of more than 145,000 People’s Liberation Army Daily articles published between January 2018 and June 2026, from which I draw a themed subset of roughly 57,000 that engage AI, command, or party control, the number of those explicitly asserting the human commander’s final authority over AI-enabled operations rose from 45 in 2018 to 153 in 2025, a 3.4-fold increase. Nor has this message been crowded out by growing enthusiasm for AI. Over the same period, AI-related coverage expanded dramatically, from 967 articles in 2018 to 2,513 in 2025, while the subset linking AI to command roughly tripled. The more the paper writes about intelligent warfare, the more often it insists that a human being remains at the end of the chain (see Figure).A word on method, since much of what follows rests on these counts. I searched the full text of each article, not just the headlines, using Chinese-language terms related to AI, command and decision-making, and party control. I then used combinations of keywords and nearby phrases to identify the themes discussed here, such as whether an article said that commanders retain final authority or raised concerns about fabricated data. These categories were not assigned by hand or by an automated classifier. An article counted as affirming commander authority when it explicitly placed the final decision, or responsibility for it, with a human commander rather than a machine, using language such as “the commander decides,” “humans bear final responsibility,” or “human-in-the-loop.” Because this is a keyword-based method, the numbers should be understood as showing how often the paper uses certain ways of framing an issue, rather than as a close interpretation of every individual article.The substantive message is remarkably consistent. However far AI advances, one 2018 article argues, the human remains the “leading factor” in war. AI is presented as an assistant: It processes information, generates and screens options, and supports commanders, but it does not make the final decision. Even the most advanced information systems, another article insists, cannot replace human judgment. When the paper discusses autonomy, it does so through the familiar progression of “human-in-the-loop,” “human-on-the-loop,” and “human-out-of-the-loop,” in which commanders delegate bounded autonomy while retaining the authority to intervene. The justification is strikingly professional rather than political. Only a human commander, the paper argues, can bear the risks and responsibilities of war, something no algorithm, however capable, can assume.The nature of the People’s Liberation Army Daily, however, also limits what this data can establish. The newspaper is the official organ of the Central Military Commission rather than an independent professional journal, and nothing appears in its pages without institutional authorization. Accordingly, its contents cannot reveal what individual officers privately believe, and I do not claim that it does. What they can establish is the position that the institution has repeatedly authorized, endorsed, and declined to revise even as its enthusiasm for artificial intelligence expands. Read this way, the insistence on human command authority is evidence not of institutional sincerity but of institutional commitment. The same standard should also govern what the paper does not say, and I apply it symmetrically below. Silence is not proof that the institution has failed to recognize a problem. It is evidence about the boundary of what the institution permits itself to state.Figure. Rising attention to AI in the People’s Liberation Army Daily, 2018 to 2025. Left: articles whose text mentions AI, intelligentization, or unmanned systems (967 to 2,513). Right: the subset of AI-and-command articles that explicitly assert the human commander’s final decision authority and responsibility (45 to 153). The counts reported here are computed on a themed subset of roughly 57,000 of those articles, selected by Chinese-language keywords for engagement with AI, command and decision, or party control. The register classification and the article-level counts underlying the figure are reproducible and available on request. A Commission That Cannot MeetSince 2017, the Central Military Commission has consisted of seven members: a chairman, two vice chairmen, and four members, including the chief of the Joint Staff Department and the directors of the Political Work and Equipment Development departments. Today, only two remain. One vice chairman and three members are gone, including the chief of the Joint Staff Department, the officer responsible for operational planning. The sole surviving vice chairman, Zhang Shengmin, spent his career in discipline inspection rather than command. With only two members left, the commission cannot function as a party committee. Analysts have likened the situation to firing all but one member of the U.S. Joint Chiefs of Staff, an arrangement under which no modern military could realistically operate.The hollowing out extends well beyond the commission itself. After Miao Hua was removed from his party and military posts in October 2025, the Political Work Department was left under a deputy director. Only the Eastern and Central Theater Commands are now led by full generals. Elsewhere, lieutenant generals occupy posts that normally carry higher rank. Across the Chinese military, acting and provisional appointments have become increasingly common. By one estimate, barely one-fifth of the military’s key leadership positions are permanently filled, while nearly half are occupied on an interim basis.Xi could reverse this quickly if he chose. A single plenary session of the Central Committee would be enough to reshuffle the commission through normal party procedures, and such sessions have been held routinely in the past. That he has not done so suggests two possibilities.Either he considers the damage to military readiness an acceptable price for political reliability, or he views the commission as increasingly ceremonial now that the Chairman Responsibility System channels operational authority through him and his personal staff. The two explanations differ, but they point toward the same conclusion: Prolonged vacancies at the apex of the military appear acceptable to Xi. What should register as an institutional crisis has instead become a sustained condition.What these vacancies mean has already been debated in these pages. Joel Wuthnow argues that the disruption is temporary, and that the greater danger comes in the late 2020s, when better-credentialed commanders arrive with less standing to push back against Xi’s war optimism. Rob Pierce argues that the purges are structurally inevitable in a system that perpetually generates reasons to purge, and predicts that fear of visible failure will produce scripted exercises and inflated readiness reporting. Christopher Nye argues that destroying the military’s informal networks will produce atomization rather than professionalization, leaving Xi to decide in an information environment divorced from reality. What none of these accounts examine is what the military itself says about any of this.Why Zhang Youxia and Liu Zhenli were removed remains uncertain. Some analysts have suggested that both men clashed with Xi over the pace of military modernization, arguing that genuine joint-operational capability cannot be built on political timelines. Whether or not that interpretation is correct, one fact is difficult to escape: With their departure, the capability gap they represented has not been filled.Can AI Rescue a Military with a “Weak and Inexperienced” Officer Corps?Here, the relationship between the purges and AI comes into focus. In Foreign Affairs in March 2026, Sam Bresnick, Emelia Probasco, and Cole McFaul drew on thousands of Chinese military procurement requests to argue that China’s intelligentization drive is further along than most observers assume. Their evidence is compelling. The Chinese military is prototyping AI across targeting, cyber operations, undersea tracking, and decision support at a pace and scale that should concern Washington. Among their conclusions is that China’s political and military leadership does not fully trust the People’s Liberation Army’s chain of command and may therefore look to AI decision-support systems to compensate for what they describe as a “weak and inexperienced” officer corps.Procurement documents reveal what the Chinese military is buying. Its newspaper reveals what it tells itself. On this point, the two diverge.Within the People’s Liberation Army Daily corpus, the substitution thesis receives almost no support. Whenever the newspaper addresses the allocation of command authority, it insists, consistently, and from the earliest years of the AI boom rather than a recent hedge, that the commander retains final authority and bears responsibility. AI is assigned a limited role: processing information, generating options, and assisting decision-making, but never replacing the commander. Equally striking is how the argument is framed. The defense of human command is made overwhelmingly in operational rather than political language. Articles invoke the commander, the battlefield, and the chain of command far more often than party leadership or political control. In my corpus, operational framings outnumber party-control framings by well over 20 to one.That ratio, however, should not be overinterpreted. In the Chinese military, professional responsibility and party control are not separate spheres. Arguments expressed in operational language can still perform political functions. Defending the commander’s prerogatives is also a way of defending the standing of the professional officer corps against encroachment — whether by algorithms or by political actors. The count therefore measures the language through which the case is made, not the underlying purpose it ultimately serves.The distinction nevertheless matters. The People’s Liberation Army Daily does not argue that commanders are indispensable because they embody party authority. It argues they are indispensable because responsibility for war cannot be delegated to a system that cannot itself be held accountable. If AI is substituting for weakened commanders, that is not a position that the institution has publicly endorsed. It is one being imposed on it despite the institution’s stated stance. That is the first reason to doubt that the substitution strategy will succeed.Bresnick and his colleagues provide the substitution thesis and the broader worry about a corrupted information environment. What the corpus adds is different in kind: evidence that the Chinese military’s own discourse rejects substitution, and a turn of the data problem inward, from an adversary’s manipulation of external data to the corruption of the military’s reporting chain by the purges meant to secure it.This distinction changes how we read the purges. If the substitution thesis had standing inside the People’s Liberation Army, the removal of its most seasoned officers would be far less consequential. Yet for nearly a decade, the institution’s own newspaper has argued the opposite: Experienced commanders remain indispensable to military decision-making. The same leadership that removed those commanders is now investing heavily in the systems that its military professionals insist cannot assume their authority. That tension does not prove the substitution strategy will fail. It does, however, suggest the central obstacle is not technological capability. It is the erosion of the very human command structure that the Chinese military continues to describe as irreplaceable.The Machines Will Eat What the Army Feeds ThemBut suppose Xi proceeds anyway. Substitution then collides with a second obstacle, and this one is not a matter of institutional preference. That the purges would corrupt the flow of information upward has already been observed in these pages: Pierce anticipated inflated readiness reporting as a consequence, and Nye argues the atomized officer corps has stopped transmitting honest assessments. Both accounts end with a misinformed leader. The chain does not end there. AI-enabled decision-making depends on the data fed into it and on the honesty of the reporting chain that produces it. If an organization cowed by purges comes to pass upward only what the leader wants to hear, that distortion flows into the very systems meant to compensate for human misjudgment. The pathology by which personal rule corrodes the information environment is not dissolved by adopting AI. If anything, it is amplified, dressed in the guise of automated certainty.This problem is not general to all AI applications. It is concentrated in systems that rely on human reporting. Systems that ingest human reporting, such as decision-support tools that draw on readiness assessments, logistics data, personnel records, and other bureaucratically generated information, are directly exposed. If those inputs reflect performative compliance rather than truthful reporting, the model necessarily inherits those distortions. By contrast, systems trained largely on sensor data, such as targeting, computer vision, or undersea tracking, are far less vulnerable to this particular pathology, because an inflated readiness report is not among their principal inputs. The argument therefore does not apply uniformly across the entire intelligentization drive. Rather, it is most consequential for precisely the applications the leadership seems most drawn to, the ones meant to compensate for a command layer it no longer trusts.The corpus lets us ask whether the Chinese military sees this coming. The answer is that it sees half of it. The People’s Liberation Army Daily discusses the data problem intensively and with real technical sophistication: Some 340 articles link artificial intelligence to training data, data quality, sampling, or data governance, rising from nine in 2018 to 77 in 2025, with 71 more in the first half of 2026 alone. Articles warn that a model may misallocate fire to secondary targets because its training data was skewed, that narrow training sets produce a “high score, low ability” trap in evaluation, and that algorithmic performance depends directly on the volume and quality of what the system is fed. The paper also discusses falsification, and increasingly so: 数据造假, data fabrication, appears six times in all of 2025 and 24 times in the first six months of 2026, while terms for false and concealed reporting run at 50 to 70 articles a year.But the two conversations do not meet. Across roughly 57,000 AI- and command-related articles over eight and a half years, the paper discusses AI and data integrity side by side in well over a hundred pieces. Yet in none of them does it connect the political incentive to misreport, created by a purge campaign, to the degradation of the data on which military AI would depend. The link that matters, from cowed reporting to corrupted training data, is simply never drawn.This absence is not an artifact of a single search. I looked for the connection under several different formulations, pairing the vocabulary of false reporting with AI, training data, poisoned data, reporting pressure, and related concepts. The resulting co-occurrences only concern civilian statistical fraud, adversarial data manipulation, or technical flaws in datasets. None addresses the possibility of the military reporting under pressure corrupting the information fed into AI.The Missing ConversationThe blind spot is therefore quite specific. The institution is not naïve about corrupted data. It has developed a sophisticated vocabulary for discussing bias, sampling, representativeness, and data governance. What it cannot name is the institutional source of those problems. Bias, sampling, and simulation fidelity are engineering problems, and the People’s Liberation Army Daily discusses them constantly. An officer who overreports his unit’s readiness because he has watched his superiors disappear is a political problem. That conversation never appears.The regime itself supplies the evidence that this is the half that matters. The new disciplinary regulations the Central Military Commission issued at the end of 2025 warned against “false war-preparedness” and “fabricated training results,” alongside incomplete implementation of the Chairman Responsibility System. The leadership, in other words, knows that units under pressure perform readiness rather than build it, and it has written the warning into military law. What neither the regulations nor the newspaper will say is what this means for the machines: Every fabricated exercise result and inflated readiness report becomes training data, and the AI meant to correct for unreliable human judgment will be learning from its lies.It is important to be precise about what this silence can and cannot mean, because two readings are available and they are not equivalent. The weaker reading is that the connection is unpublishable. No one can argue in the Central Military Commission’s newspaper that the Chairman’s purges encourage officers to lie and thereby degrade the data supporting military AI. On this reading, the silence tells us nothing about whether Chinese military professionals see the problem, only that they cannot say so in print. The stronger reading is that the silence reflects genuine institutional non-recognition, that the connection has not been drawn because it has not been seen. Open sources cannot settle which is true, and the argument here does not rest on the stronger one.But the weaker reading is not a weak finding. A military that cannot name a problem in its own press cannot work on it there either: it cannot debate the problem, circulate a corrective doctrine, train against it, or build the habits that would contain it. What is unspeakable in the venue where an army thinks aloud is not something that army can address. And the reason this problem is unspeakable is that naming it would mean identifying the leader’s own campaign as the source of the corruption. The purges create the incentive to misreport — the same political order that creates the incentive forecloses discussion of what it costs.The circle is therefore complete. The tighter the political control that makes honest reporting dangerous, the poorer the data on which military AI depends. The poorer the data, the less AI can deliver the compensation Xi appears to be seeking.The purges and China’s turn toward AI are therefore not separate developments. The former reflects distrust of human commanders. The latter reflects hope that machines can compensate for what those purges have removed. AI may ultimately be Xi’s wager: a bet that technology can offset the institutional damage produced by his own system of political control.Jaehwan Lim is a professor of international relations at Aoyama Gakuin University, where his research focuses on authoritarian politics and contemporary China, including Chinese elite politics and political economy. His current work applies qualitative and quantitative text analysis to Chinese media and policy corpora. His research has appeared in the China Quarterly, China Economic Review, the Developing Economies, International Economics, and Journal of Contemporary China, among others.Image: COP PARIS via Wikimedia Commons