Palantir Maven’s 'War of Speed': Structural Warnings from the Minab Misdirected Strike
Executive Summary
The February 2026 airstrike on an elementary school in Minab, Iran, was a systemic incident that occurred when the Palantir Maven system classified targets without the verification authority to filter out outdated information. The pressure of processing approximately 13,000 targets over 38 days structurally induced automation bias, rendering human verification hollow. While Washington denies the possibility of war crimes, labels related reporting as betrayal, and refuses to accept responsibility, Iran is expanding the diplomatic front to include the accountability of US allies through a letter to the UN Security Council. Given the structure in which Maven-family systems are standardized and supplied to allies, this flaw could spill over into South Korea’s ROK-US combined intelligence systems and its future adoption of AI-enabled Command and Control (AI-C2). South Korea must move away from the passive acceptance of standards and secure the rights to data provenance disclosure and algorithmic verification at both contractual and institutional levels.
I. Situation Analysis
AI Military Targeting Controversy: Situation Analysis of Civilian Casualties in the Minab Airstrike in Iran
1. Background and Progress of the Issue
On February 28, 2026, the United States and Israel launched attacks on Iran. Supreme Leader Ali Khamenei was killed in the airstrikes, and Iran's nuclear and military facilities were also hit [2][3]. Pakistan subsequently intervened to mediate, and a June agreement temporarily halted large-scale combat. However, both sides have since continued to exchange sporadic attacks, citing violations of the agreement [2][3].
During the first 38 days of the war, an AI-based targeting system was operational, identifying approximately 13,000 targets during this period [3]. One of these targets was an elementary school in the Minab region of southern Iran. On February 28, two Tomahawk missiles struck the school [1], killing more than 150 people, including 123 children [3]. The *Tehran Times* characterized this airstrike, along with an attack on a civilian area in the Ramrad region, as 'incidents of potential war crimes' [14].
The Maven system developed by Palantir was utilized in the targeting process [1][3]. According to a Bloomberg investigation, the system classified the school as a target based on outdated information, while human verification procedures had been scaled back [1][4]. Adnane Kaab, a former senior officer of the Royal Moroccan Air Force writing for the Moroccan media outlet *TelQuel*, pointed to a different essence of the incident. He assessed that 'AI did not make the decision. AI merely changed the speed and structure of the decision' [1].
2. Current Situation
More than six months after the incident, the dispute over responsibility shows no signs of abating. Iran sent a letter to the United Nations Security Council requesting an investigation into the role that European territories, military bases, and infrastructure played in supporting US military operations against Iran [13]. Through this, the Iranian side made clear its intention to hold European nations legally accountable under international law [13].
The reaction within the United States is starkly different. The White House immediately denied the UN fact-finding mission's assessment of potential war crimes [3]. Secretary of Defense Pete Hegseth characterized reporting on the war with Iran as 'a betrayal of the nation,' labeling the journalists responsible for those reports as anti-American actors [11]. Instead, Washington formalized its position that the military operation demonstrated 'unparalleled technological superiority' [Deleted Source].
*Foreign Affairs* highlighted the structural flaws of the system, pointing out that the Maven system, lacking verification authority, likely misclassified the Iranian school as a target after receiving poor-quality input data [4]. The same article warned that when the speed of targeting becomes excessively fast, the risk of automation bias—where humans completely delegate judgment to algorithms—increases [4]. A similar pattern was repeated later. This spring, a case was reported where an interdiction and seizure operation against a Chinese vessel was prepared based on an intelligence report co-authored by an AI chatbot, only to be aborted at the last minute after fighter jets had already scrambled [8]. In this case, the source of the error was not deception by the adversary, but rather the US military's own intelligence production process [8]. The *Tehran Times* went a step further, citing a *Time* magazine report that the Trump administration virtually delegated military decision-making to a large language model (LLM) following a closed-door session with Grok AI in December 2025 [15].
3. Key Actors and Positions
The US Administration and Department of Defensemaintain a stance that emphasizes the technical achievements of the operation. The White House did not accept the findings of the UN fact-finding mission [3]. Secretary of Defense Hegseth responded aggressively, defining media reports as security threats [11]. This is interpreted as an attitude that prioritizes securing the legitimacy of the operation over acknowledging responsibility for civilian casualties.
Palantiris a key party to this incident as the provider of the Maven system. The issues of outdated data and scaled-back human safeguards identified by the Bloomberg investigation are directly linked to the structural limitations of the targeting algorithm designed by the company [1][4]. Given that the same family of systems is standardized and supplied to multiple allies, concerns have also been raised that this error could spill over into the operational systems of other nations [3].
The Iranian Governmentis pursuing a strategy of internationalizing accountability for the damage. A prime example is its request to the UN Security Council to investigate the role of Europe [13]. Iranian media outlets, including the *Tehran Times*, continue to report on the Minab airstrike using a war crimes frame, focusing on highlighting the failure of US intelligence management and its chaotic decision-making process [14][15].
Security Experts in the Middle East and North Africa (MENA) Region, including Morocco,view this incident not as a mere technical glitch, but as a fundamental shift in the structure of military decision-making itself. As Kaab analyzed, their assessment is that AI intervention did not replace human judgment but rather compressed the time available for making decisions [1]. This implies that accountability should be sought not in the algorithm itself, but in the command structure and institutional design that operate it.
International Governance Organizations, including SIPRI,are conducting discussions on establishing military AI norms through cooperation between states and the defense industry. A closed-door dialogue hosted by SIPRI in September was attended by officials from various ministries of defense and foreign affairs, as well as representatives from the International Committee of the Red Cross (ICRC), NATO, and the United Nations Institute for Disarmament Research (UNIDIR) [5]. However, these discussions have not yet led to an agreement with concrete binding force.
4. Key Issues
The first issue is the trade-off between speed and verification. The argument that the processing speed of identifying 13,000 targets in 38 days structurally compressed human verification time is gaining traction [3][4]. The second is the diffusion of responsibility. The entanglement of Palantir as the algorithm designer, the US Department of Defense as the operator, and the poor quality of the data input process has created a structure in which it is difficult to pinpoint legal liability [1][4]. The third is the risk of spillover to allies. With the same Maven system being supplied as a standard to multiple friendly nations, concerns are rising that this error may not be a one-off accident but a repeatable structural flaw [3]. The fourth is the effectiveness of international accountability. Despite Iran's petition to the UN and the UN fact-finding mission's assessment of potential war crimes, the US administration refuses to accept these findings, leaving a persistent gap between international legal accountability and actual political responsibility [3][13].
II. In-Depth Analysis
AI Military Targeting Controversy: In-Depth Analysis of Civilian Casualties in the Minab Airstrike in Iran
1. Root Cause Analysis
The primary cause of the misdirected strike on the Minab school was a defect in the intelligence itself. *Foreign Affairs* pointed out that the Maven system likely classified the Iranian school as a target without being granted the authority to filter out poor-quality input data [4]. This indicates that verification functions were already restricted at the system design stage. The Bloomberg investigation also highlighted the same point, concluding that it was the result of combining outdated data with scaled-back human safeguards [1].
The core of the problem lies in speed. During the first 38 days of the war, the AI targeting system identified approximately 13,000 targets [3], averaging over 300 targets filtered per day. At this processing speed, it is virtually impossible for human verifiers to individually reconfirm the source and reliability of information for each target. The automation bias warned of by *Foreign Affairs*—where humans completely delegate judgment to algorithms—stems precisely from this pressure of speed [4]. The diagnosis by Moroccan strategic analyst Adnane Kaab shares this context. He assessed that 'AI did not make the decision. AI merely changed the speed and structure of the decision' [1]. This suggests that rather than being the active agent of misjudgment, AI is a mechanism that structurally compresses the time available to filter out such errors.
The second cause is excessive trust in private corporations. The *TelQuel* report explicitly identified 'excessive trust (*confiance excessive*)' in Palantir's Maven system as the background of this incident [1]. This implies that by relying on a private technology company's platform for the core military function of targeting, the Department of Defense lost its independent capability to verify the platform's data processing methods and algorithmic logic. This should be viewed not as a simple technical glitch, but as an issue of the distribution of authority between the military and defense contractors.
2. Structural Context
Political Structure: Institutionalization of Evasion of Responsibility
The White House immediately denied the UN fact-finding mission's assessment of potential war crimes [3]. Secretary of Defense Pete Hegseth characterized reporting on the war with Iran as 'a betrayal of the nation' and labeled the reporting journalists as anti-American actors [11]. This response is not an isolated statement but a pattern. Instead, Washington formalized its position that the military operation demonstrated 'unparalleled technological superiority' [Deleted Source]. This points to an underlying political calculation that acknowledging flaws in the AI targeting system would undermine the very legitimacy of the war effort.
Iran is exploiting this gap on the international stage. It sent a letter to the UN Security Council requesting an investigation into the role that European territories, military bases, and infrastructure played in supporting US operations against Iran [13]. The *Tehran Times* grouped the Minab school airstrike and the attack on the Ramrad civilian area together, defining them as 'incidents of potential war crimes' [14]. The focus of Iranian diplomacy is shifting away from the debate over internal US technical flaws toward the legal accountability of the entire allied network that enabled the US military operations.
Economic Structure: The Defense-AI Enterprise Complex
Palantir, which supplied the Maven system, is a publicly traded private corporation. The Department of Defense entrusting core targeting functions to the commercial platform of such a private company means that a critical link in military decision-making can be influenced by corporate commercial interests and product roadmaps. This concern formed the backdrop of the closed-door dialogue hosted by SIPRI in September, where officials from ministries of defense and foreign affairs, the ICRC, NATO, and UNIDIR gathered to discuss state-industry cooperation for lawful military AI development [5]. The problem is that this dialogue has not yet translated into binding norms.
Security Structure: Spillover Risks of Allied Standardization
Systems like Maven are not tools confined to the United States alone. Within a structure where the same AI-C2 standard is exported to and shared among multiple allies, the core structural risk is that an error in a single data layer or a flaw in verification procedures can spill over to all nations sharing that standard. For a middle power like South Korea that adopts US-led AI-C2 standards, the possibility of replicating the same errors cannot be ruled out if it accepts the system without independent control over targeting logic and data provenance.
3. Historical Precedents and Comparison of Similar Cases
The closest precedent to this incident is the case of a misdirected operation against China that occurred during the same war. This spring, the US military prepared an interdiction and seizure operation against a Chinese cargo ship in Middle Eastern waters based on an intelligence report co-authored by an AI chatbot [8]. The report concluded that the vessel was carrying components related to a nuclear weapons program, which was not true [8]. The operation was aborted at the last minute while fighter jets were already airborne [8]. The core of this case is that the source of the error was not adversary deception, but the US military's own intelligence production process [8]. A flawed conclusion gained institutional credibility by passing through a standardized intelligence report format, and in that state, it was rapidly circulated through the chain of command [8].
Placing the Minab school airstrike and this case side by side reveals a common structure. In both cases, information generated or assisted by AI passed through standardized procedures without verification. In both cases, a point for human verifiers to intervene virtually did not exist or existed only pro forma. The only difference is that the Minab incident led to an actual strike and large-scale casualties, whereas the Chinese vessel case was aborted at the last minute, preventing damage. This indicates that the same structural flaw merely produced different outcomes, rather than serving as evidence that the system itself had improved.
The *Time* magazine report relayed by the *Tehran Times* raises an even more fundamental issue. It alleges that President Trump held a closed-door session at the White House with Elon Musk in December 2025 to consult Grok AI on military judgments [15]. While the factual accuracy of this report requires cross-verification given its source in Iranian state media, if true, it implies that reliance on AI has penetrated beyond the operational level into the strategic decision-making stage at the highest echelons of the military chain of command.
Since the Cold War, the issue of automation bias has been a recurring theme in military technological history. The 1983 Soviet early warning satellite false alarm incident and the Patriot missile malfunctions during the Gulf War demonstrated the risks that arise when humans fail to sufficiently question machine-provided information. However, where the Minab incident qualitatively differs from previous cases is that the time to detect and halt errors was structurally compressed from the very stage of system design. The automation bias pointed out by *Foreign Affairs* must be viewed not as an accidental mistake, but as an inherent risk embedded in the system's design philosophy [4].
4. Key Variables Shaping Future Developments
The first variable is the international binding force of the UN fact-finding mission's investigation results. The UN mission has already raised the possibility of war crimes [3]. However, with the White House immediately denying this [3], the likelihood of practical sanctions or accountability being imposed on the United States, a permanent member of the UN Security Council, remains low. Iran's separate targeting of Europe's role [13] can be interpreted as a detour strategy reflecting its recognition of the limitations of directly holding the US accountable.
The second variable is whether a regulatory framework will be established for private defense AI companies, including Palantir. The state-industry dialogue hosted by SIPRI[5] remains in the discussion stage. Whether this dialogue leads to binding certification standards or verification obligations, or remains merely declaratory, will shape the future direction of military AI governance.
The third variable is the pace at which the US-led AI-C2 standard spreads to allies. The extent of the risk that similar errors will be replicated in other theaters depends on how much independent control over data layers and verification procedures US allies and partners, including South Korea, can secure during the process of adopting this standard. This is also a matter of how much leverage individual nations can exert at the technology transfer negotiation table.
The fourth variable is the trajectory of the political debate over responsibility within the United States. In a situation where the Secretary of Defense defines war-related reporting as 'betrayal' [11], the very environment for whistleblowing or further disclosures is suppressed. Whether investigative reporting by outlets like Bloomberg or *Time* magazine continues, or whether the administration's information control proves effective in dampening the debate, will determine the political longevity of this issue.
III. Recommended Policy Responses
AI Military Targeting Controversy: Civilian Casualties in the Minab Airstrike in Iran — Comprehensive Policy Responses
1. Comprehensive Assessment
The misdirected strike on the Minab school is not a mere technical failure. It is a systemic incident in which three layers overlap: the distribution of authority between the military and private defense contractors, wartime intelligence verification procedures, and the adoption of allied AI-C2 standards. The Palantir Maven system is not used by the US Department of Defense alone [3]. Because the same family of systems is supplied to multiple allies in a standardized format, the flaws revealed in this incident do not end as the problem of a single nation [3]. If South Korea accepts this structure as is during the process of integrating ROK-US combined intelligence systems or adopting future AI-C2, it is effectively accepting the very pathway through which identical errors can spill over [3][8].
Washington's pattern of response has already been established. The White House denied the UN fact-finding mission's assessment of potential war crimes [3], and the Secretary of Defense characterized related reporting as 'a betrayal of the nation' [11]. This indicates that the United States has little incentive to officially acknowledge the structural flaws of its AI targeting system. For South Korea to address this issue, rather than waiting for voluntary corrective action from the US, it must choose to institutionally secure its own independent verification rights within the combined intelligence systems.
2. Policy Recommendations
The core direction must shift from the 'passive acceptance of AI-C2 standards' to 'securing active control over data layers and verification procedures' [3]. This does not mean rejecting the alliance itself; rather, it is an effort to clarify the scope of responsibility that South Korea will bear within the alliance.
The Ministry of National Defense and the Joint Chiefs of Staff must comprehensively identify all points where AI-generated information enters the ROK-US combined intelligence systems. As demonstrated by the case of the misidentified Chinese vessel this spring, errors can arise not only from adversary deception but also from internal intelligence production processes [8]. They must inspect the very pathway through which flawed conclusions gain institutional credibility by passing through standardized report formats [8].
3. Short-Term Action Plan (3–6 Months)
The Ministry of National Defense (MND) should conduct a comprehensive survey of the scope of application of AI-assisted targeting and intelligence analysis tools within the ROK-U.S. combined intelligence system. This must be preceded by mapping out a flowchart that details at which stage AI intervenes and at which stage humans perform the final verification.
There is a need to establish a separate verification channel within the Joint Chiefs of Staff (JCS) for AI-generated intelligence. Rather than allowing standard report templates to pass through as they are, a procedure must be mandated to explicitly indicate whether AI assistance was utilized [8].
The Ministry of Foreign Affairs (MOFA) should continuously track the findings of the UN fact-finding mission on the Minab incident and the contents of Iran's letter to the Security Council [3][13]. The possibility cannot be ruled out that the discussion raised by Iran regarding the responsibility of allies could expand to U.S. Forces Korea (USFK) bases or South Korea's military cooperation with the United States.
4. Medium-Term Action Plan (1–2 Years)
When introducing Maven-class or similar AI-C2 systems, the Defense Acquisition Program Administration (DAPA) must specify the obligation to disclose data sources and the right to verify algorithmic logic in contract clauses. The issue of "excessive trust" pointed out by TelQuel [1] is repeated when the military accepts the black boxes of private companies without verification.
The Agency for Defense Development (ADD) and the JCS should pursue the development of an independent verification module for AI-assisted targeting systems. The goal is to establish a dual system where U.S. standards are adopted but South Korea can re-verify them using its own data.
Similar to the state-industry dialogue led by SIPRI, South Korea also needs to establish a permanent military AI governance consultative body involving defense companies, the military, and academia [5]. This model, which involved the International Committee of the Red Cross (ICRC), NATO, and the UN Institute for Disarmament Research (UNIDIR), can function as an institutional mechanism to reduce the military's dependence on private technology companies [5].
5. Long-Term Action Plan (3 Years or More)
South Korea should participate as a member of the middle-power group in the process of shaping international norms on military AI targeting. This incident went beyond the issue of the responsibility of a single nation, the United States, and revealed a common risk for all allies adopting AI-C2 standards [3]. Working with middle powers that share this awareness to put the standardization of verification procedures on the agenda is a long-term task.
The concept of data sovereignty must be institutionalized across the entire defense AI procurement system. The goal is to avoid a structure where core military functions are dependent on a specific vendor's platform, and to simultaneously secure multiple supply lines and independent verification capabilities.
6. Monitoring Indicators and Trigger Points
Whether the UN fact-finding mission releases its final report on the Minab incident and the contents of that report serve as the primary indicator. If a determination of war crimes is formalized, it could lead to changes in the United States' policy on exporting AI-C2 standards to its allies.
The progress of Iran's request to the Security Council to investigate the European role is also a subject to track [13]. If this discussion materializes into holding specific allies accountable, South Korea must preemptively review the possibility of similar issues being raised regarding its own military cooperation structure with the United States.
Whether the U.S. Congress or the Department of Defense inspector general launches an official investigation into the Maven system is another trigger. Although the White House currently maintains a consistent denial [3], the initiation of an independent congressional investigation could lead to design changes in the system itself.
The recurrence of similar accidental bombings or misidentification incidents is also a key indicator. If incidents stemming from the same automation bias structure are repeated—such as the case of misidentifying and intercepting a Chinese vessel this spring [8]—it would serve as evidence supporting that this is not a one-time glitch but a structural problem across the entire system.
7. Summary and Conclusion
The accidental bombing of the Minab school is an incident that revealed how AI is changing the speed of war and decision-making structures [1]. The combination of three factors—outdated data, reduced human verification, and excessive trust in private companies—resulted in more than 150 civilian casualties [1][3]. Washington chose denial over acknowledging responsibility [3][11], and Iran is expanding this into a debate over the responsibility of allies [13].
The task for South Korea is clear. In a structure where the same line of AI-C2 systems is standardized across multiple allies, South Korea must not remain in a passive position of merely accepting U.S. decisions [3]. Securing institutional rights to independent verification within the combined intelligence system, access to data sources, and the authority to audit algorithmic logic is the most concrete task this incident has left for South Korea [3][8].
References
[1] [TelQuel] [Tribune] Guerre et IA : quand la machine change la chaîne de décision
[4] [Foreign Affairs] How Iran Changed War
[7] [Council on Foreign Relations (CFR)] What Questions Should Leaders Ask Before Going to War?
[10] [Brookings Institution] The Iran war and America’s widening strategy gap in Asia
[11] [Australian Financial Review] Wong confirms two Australians safe after Flydubai attack
[13] [Tehran Times] Iran urges UN to examine European role in US military operations against Iran
[14] [Tehran Times] Rubio invokes ‘blackmail’ to hide US crimes in Iran
[15] [Tehran Times] Time magazine exposes recklessness at the heart of Trump’s warmongering
[16] [Hürriyet Daily News] UN says strikes kill 10 civilians in Afghanistan
[18] [Le Monde] 'Factory of targets' and 'collateral damage': Israeli soldiers describe Gaza methods
[19] [Cyprus Mail] Iran readies harder retaliation if attacked as diplomacy faces long odds
[21] [DW (Deutsche Welle)] Yemen: Houthis report deadly Saudi strikes on Taiz market
[22] [Tehran Times] Report: Eight US Marines wounded in Iranian strike near Strait of Hormuz
[23] [Wired] AI Is Making a Mess of Nurses’ Schedules. They Say It’s a Safety Issue
[24] [La Repubblica] Live: Attempted attack on RAF base: British-Iranian suspect arrested
*This text is an AI translation of an original written in Korean. Some translations or nuances may be inaccurate.
This report is an in-depth analysis planned by an EAI researcher, grounded in sophisticated AI-assisted research, and finalized by the EAI researcher.
How can I help you?