An Iran-linked entity recently employed Anthropic’s Claude artificial intelligence platform to compile publicly available data on U.S. Navy forces and subsequently develop sophisticated targeting recommendations, a revelation detailed in a new threat intelligence report from the AI company itself. Anthropic confirmed it successfully detected and disrupted this illicit operation, leading to the immediate banning of the associated account and the sharing of pertinent information with government authorities, underscoring the escalating national security challenges posed by the dual-use nature of advanced AI technologies.
The disclosure by Anthropic serves as a stark reminder of a burgeoning concern for military strategists and policymakers worldwide. The very AI systems being rapidly developed and deployed to process vast quantities of information for legitimate defensive and intelligence-gathering purposes can, in the hands of adversaries, be weaponized to transform disparate public data into actionable intelligence, thereby compressing complex intelligence workflows that once demanded extensive human expertise and significant time.
The Modus Operandi: Automating Intelligence for Military Planning
According to Anthropic’s investigation, the Iranian-linked actor meticulously utilized Claude to construct an automated pipeline designed to assemble comprehensive targeting handbooks. This sophisticated system drew upon a diverse array of open-source intelligence (OSINT) inputs, including ship and aircraft transponder identifiers, commercially available satellite imagery, and publicly circulated photographs of U.S. military personnel. Beyond mere data aggregation, the actor also leveraged the advanced analytical capabilities of the AI model to conduct in-depth research into potential vulnerabilities within critical shipboard systems. These included maritime satellite terminals, widely used Cisco communications equipment, and various industrial-control products, indicating a clear intent to identify exploitable weaknesses in naval infrastructure.
This particular incident, one of several highlighted in Anthropic’s extensive 154-page report spanning activity between December 2025 and August 2026, illuminates not merely that state-linked actors are exploring AI, but rather how commercially available generative AI systems can dramatically streamline and accelerate portions of an intelligence workflow that traditionally necessitated considerable specialist effort and resources. The report explicitly clarified that Claude was not employed to autonomously initiate an attack. Instead, its role was confined to the collection, organization, and analysis of information, automating crucial open-source intelligence tasks, and ultimately producing material directly supportive of military planning and decision-making processes. The initial findings of this significant breach were first reported by The Wall Street Journal, before Anthropic provided a more granular account in its latest report on the misuse of its AI models.
The Global Proliferation of AI in Modern Warfare
Modern targeting has increasingly become a complex data-processing challenge, requiring military forces to sift through and interpret an overwhelming volume of information from diverse sources. This includes high-resolution satellite imagery, continuous drone video feeds, electronic signals intelligence, intercepted communications, and countless other data points, all at a pace that far exceeds the capacity of human analysts working in isolation.
The United States has been at the forefront of this technological shift for years. A prime example is the Pentagon’s Project Maven, initiated to harness machine learning and computer vision for the rapid processing of imagery and the identification of objects of military interest. Early iterations of Maven systems were designed to scan vast quantities of visual data, flag potential targets, and seamlessly integrate this information into existing targeting workflows, always with human analysts retaining the critical responsibility for validation and final decision-making. The U.S. Army has consistently expanded these capabilities, demonstrating in a 2026 exercise the Maven Smart System as an AI-enabled tool capable of processing thousands of hours of surveillance video to automatically identify and flag objects of interest. This illustrates a clear trend: AI-assisted targeting is not a capability exclusive to any single nation but is becoming an integral component of military modernization strategies across all major global powers.
Ethical Quandaries and the Indispensability of Human Oversight
The integration of AI into military targeting processes has become a particularly sensitive and ethically charged issue, especially in the wake of recent high-profile incidents. The U.S.-Israeli strikes on Iran on February 28, for instance, saw a U.S. strike hit the Shajareh Tayyebeh school in Minab, tragically killing at least 168 people, predominantly children, according to reports by The Associated Press. Subsequent investigations by The Washington Post revealed that the school had inexplicably appeared on a U.S. target list, potentially having been mistakenly identified as a military site. While questions were raised about the potential role of AI-assisted target identification in this incident, it has not been definitively established that AI directly caused the strike or independently selected the school as a target.
Similarly, Israel has openly incorporated AI-assisted systems into its military operations, particularly since the commencement of its campaign in Gaza. Reports have extensively examined systems such as "Lavender," which Israeli officials have consistently described as an analytical aid intended to support human decision-making rather than an autonomous target-selection system. In 2024, Reuters reported that the United States was scrutinizing allegations of Israel’s use of AI to identify bombing targets, with the Israeli military reiterating that human analysts remained ultimately responsible for target identification and final operational decisions. These examples, while not diminishing the gravity of Iran’s reported use of Claude, underscore a broader, systemic challenge that transcends individual nations: the critical importance of robust human oversight in AI-driven warfare and the inherent risks when complex data, potentially incomplete, misleading, or erroneous, feeds into autonomous or semi-autonomous systems.
Broader Patterns of AI Misuse: From Missiles to Surveillance
The Iranian naval intelligence operation represents just one facet of a disturbing trend documented in Anthropic’s comprehensive report. The company detailed several other concerning instances of AI misuse by state-linked and malicious actors. Users operating from northern Yemen, for example, allegedly attempted to exploit Claude’s capabilities to develop software for advanced missile systems. Anthropic confirmed that while some of these requests were successfully blocked by its internal safeguards, others managed to circumvent initial protections before the accounts were ultimately banned.
The report also brought to light operations aligned with the Chinese government, which reportedly used Claude for surveillance and profiling activities, including targeted efforts against Uyghurs and religious figures, raising profound human rights concerns. Concurrently, Russian-linked actors were documented attempting to leverage Claude to develop sophisticated autonomous drone swarms, drawing upon intelligence and data gleaned from the ongoing conflict in Ukraine. These varied incidents collectively paint a picture of a rapidly evolving threat landscape, where general-purpose AI is being adapted for a wide spectrum of military and surveillance applications.
Anthropic maintains that it successfully disrupted all the operations detailed in its report and has since significantly strengthened its detection systems and safeguards. However, these cases highlight an uncomfortable and increasingly pressing reality for the entire artificial intelligence industry: preventing dangerous misuse is far more complex than simply refusing an overtly malicious query. Anthropic’s findings indicate that sophisticated actors are deliberately fragmenting their illicit requests into individually benign tasks. This method allows them to bypass initial safeguards by building more dangerous capabilities incrementally over multiple, seemingly innocuous interactions, effectively assembling a mosaic of harmful functionalities from seemingly harmless pieces.
The Inevitable Future: AI in Warfare and the Imperative of Control
As AI systems continue to advance in their capabilities across coding, research, image analysis, and the handling of massive datasets, the military significance of these general-purpose tools will inevitably continue to grow. The central question, therefore, is no longer whether artificial intelligence will be integrated into warfare. It already is, irrevocably shaping the future of conflict.
The more consequential and ethically profound question revolves around the degree of authority humans will retain over these increasingly powerful systems that assist in finding, prioritizing, and ultimately engaging targets. It challenges societies and military doctrines to establish whether the safeguards put in place to ensure human control and accountability will be robust enough, especially when the vast amounts of data feeding these machines are incomplete, potentially misleading, or simply inaccurate. The incidents detailed by Anthropic serve as a stark global warning, compelling an urgent and widespread dialogue on the governance, ethics, and responsible development of AI in an increasingly volatile world. The future of global security hinges on our ability to navigate this complex technological frontier with foresight, caution, and an unwavering commitment to human values.
