Contents
Introduction
1) What do we mean by AI-enabled warfare?
2) What do we know about how AI is being used in the current war? 
3) What is the role of AI in disinformation in this war?
4) Why and how is AI making mistakes?
5) What does "human oversight" mean in these contexts? Aren't humans supposed to double-check AI-generated lists in order to ensure that no civilians are included?
6) Are there no legal barriers to the use of AI in warfare?
7) Does the recent Anthropic/Pentagon feud have anything to do with current events?
8) What is the impact of this war on the infrastructure that supports AI? 
9) What else has been happening in the MENA tech ecosystem since the start of the war?

Introduction

In light of the Israel-US attacks on Iran and the ensuing conflict, there has been a large body of journalism and reporting that has examined the implications of developments in AI and AI-enabled warfare, particularly in relation to the current war. This scrutiny is unfolding within a broader context of growing concern regarding the use of automated weapons systems, AI-based decision support systems, and privately contracted technologies by militaries around the world, most notably the US and Israel. Companies such as Alphabet, Anthropic, Microsoft, and OpenAI have recently been embroiled in scandals surrounding the unethical use of their technologies in the US and abroad, including during the genocide in Gaza.

As in any war, it is hard to ascertain the exactitude of information while the fighting is ongoing. On one hand, under the guise of protecting national security interests, governments and contracted companies intentionally obscure operating mechanisms, data classification and processing methods, and the nature of contracts and procurement processes. On the other hand, various state and non-state actors are increasingly using disinformation to confuse the information ecosystem and cast doubt on any updates that come out of conflict zones. This is further exacerbated by internet shutdowns, online censorship, and an increasing polarization of online media.

As a result, there are still major gaps in what is known. The private-military black-box remains one of the hardest to crack even as open-source intelligence (OSINT) becomes increasingly valuable and important.

This Q&A seeks to clarify how AI-enabled warfare and disinformation are shaping the current war, while also acknowledging the limits of what can be known with certainty. It draws on available reporting, official statements, and existing research to explain what technologies are being used, who is enabling them, how they are affecting military operations, and what they mean for civilians and critical infrastructure. More broadly, it places these developments within wider debates about accountability, opacity, and the growing role of private technology companies in war.

1)  What do we mean by AI-enabled warfare?

Defining "AI-enabled warfare" is a challenging task, as it is a catch-all term that describes various military functions. It can designate systems that provide support for, among others: intelligence processing, target nominations or recommendations, “last mile targeting”, controlling systems that are wholly autonomous throughout engagement, controlling swarms of systems or munitions acting in concert, logistics roles, the development of new tactics and doctrines. In warfare, this can include the use of autonomous drones, autonomous machine guns, and creating options for ways to strike specific targets, including necessary material and personnel deployments.

2)  What do we know about how AI is being used in the current war?

There are several documented uses of AI or AI-enabled technologies in the current US-Israel-Iran war. Their use highlights the increasingly blurred lines between private enterprises and the military itself, who now outsources many operational matters of national security to contractors.

The US

In 2024, Palantir publicly announced the integration of Maven Smart System (MSS), in conjunction with Anthropic's Claude AI and Amazon Web Services (AWS), to US intelligence agencies in order to provide decision-making support to officials. The system appears to work with three inter-linked layers of data input/processing. The US military provides data such as satellite imagery, munitions loads, and troop capabilities to Palantir's MSS. MSS is then used by the military, with Palantir support, to collate, fuse, and model this data and process it through several different analytical systems, including Claude AI, which can help interpret the information and produce target recommendations as 'user-friendly' outputs. The entire process is supported by AWS, which provides the underlying cloud infrastructure that enables storage, computing power, and scalability.

The most open example of MSS capabilities was shown in Palantir's AIPCon 9, the company’s March 2026 customer conference showcasing applications of its Artificial Intelligence Platform across sectors including defense, where a representative from the US Department of War demonstrated in real-time how MSS is used to collate and secure several different decision-making processes, including target nomination and striking. The video hints at potential systems named "Maverick" – presumably for missiles – and "SMACK," but no further data was found during research. According to Katrina Manson, investigative journalist and author of "Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare", MSS has been used to strike approximately 10 000 targets in Iran.

Israel

Israel uses systems such as "Where's Daddy?", "Lavender" and "The Gospel", first reported on by the +972 magazine, Local Call, and the Guardian. These systems rely on the nonconsensual, large-scale, and wide-reaching collection, processing, and operationalization of Palestinian data for military purposes. “Lavender” was said to identify suspected militants at scale, “Where’s Daddy?” to monitor when they were at home, and “The Gospel” to generate and rank buildings and sites for attack. While there has been no documented proof of their use in Iran or Lebanon currently, the speed and number of attacks suggest AI-enabled processes possibly similar to those used in Gaza since 7 October 2023.

Israel's Iron Dome system also uses AI to help intercept missiles, rockets, or mortars headed toward high-value targets. Local reporting from Ynet and The Jerusalem Post, as well as research by the International Institute for Strategic Studies, has suggested that the IDF also runs several other technologies and AI-enabled systems in what has been called an "information factory".

In addition to the systems outlined above and in the hyperlinked texts, ARI has found reporting on the following tools: "Wolf Pack", "Red Wolf", and "Blue Wolf", for wide-ranging surveillance and data collection on Palestinians; "Flow", which allows for the extraction of information from various datasets; "Hunter", which profiles and signals potential targets; "Depth of Wisdom", to map the tunnel network in Gaza; and "Fire Factory", which "uses data about military-approved targets to calculate munition loads, prioritize and assign thousands of targets to aircraft and drones, and propose a schedule".

3)  What is the role of AI in disinformation in this war?

Disinformation in wartime is not new. The advent of AI-generated deepfakes has made it all the more difficult to distinguish real from fabricated content. Media now circulates AI-made videos to falsify narratives via audio-visuals, while audiences are increasingly disputing the authenticity of online content.

In the present conflict, AI deepfakes have been used extensively by all sides and various individuals to falsify or exaggerate on-the-ground realities in the conflict, appealing to publics’ emotions, and pushing governments in affected states to take stronger measures to contain both the narrative and its consequent shock.

Indeed, Iran's primary use of AI appears to be the fabrication of "battle successes" to shape public perception. By generating high-fidelity deepfakes, the regime creates a digital fog of war that forces OSINT analysts to expend significant resources on verification. AI-generated satellite imagery has been used to falsely claim the destruction of US bases in the GCC. These fakes often use real Google Earth data as a base, adding AI-generated layers of fire and wreckage. Iranian agencies like SalamPix have successfully funneled AI-generated images into mainstream media. Notable outlets, including Germany’s Der Spiegel, were forced to retract images of "Iranian aircraft carriers" after forensic analysis revealed they were AI constructs.

The rampant use of AI-generated images for the purposes of propaganda by all parties has distorted debates and eroded trust, giving rise to the phenomenon of a “liar’s dividend”, where disinformation actors can avoid responsibility by disputing the authenticity of media content. As a notable example, amid circulating rumors of Israeli Prime Minister Benjamin Netanyahu’s death in March 2026, videos posted to his social media as evidence of well-being have been dismissed by online audiences as AI.

A second issue posed by AI-generated disinformation online is how it may impact military operations. While the data collection mechanisms used to process “real-time" information are heavily guarded, it is possible that they draw on online sources such as social media posts or news reporting to inform decision making. This means that, without strong checks and human verification, (dis)information risks being integrated into military datasets and subsequently treated as reliable data on which to base decisions. While there have been no confirmed public cases of this, the documented use of OSINT in conflict analysis means that polluted data environments pose a credible operational risk in the absence of robust verification mechanisms.

4)  Why and how is AI making mistakes?

The question of mistakes in AI-enabled systems needs careful framing. Zena Assaad, Associate Professor in the School of Engineering at the Australian National University, stresses that failures in such systems can be understood at the technical level, at the human level, and in the interaction between the two. The bombing of the Shajareh Tayyebeh girls' elementary school in Minab, Iran is exemplary of this tension.

The technological system that proposed the target and launched the strike is widely accepted to have worked as designed, that is to say that it produced a supposed military target based on data that was provided by humans. In this specific strike there has been no mention of a technical system malfunction, though these systems are known to have low accuracy rates, sometimes dropping as far as 30% in certain conditions such as bad weather – a phenomenon noted even outside of military uses of AI. Indeed, Anthropic itself has stated that it "currently does not have confidence, for example, that Claude would function reliably or safely if used to support lethal autonomous warfare".

These systems are also, however, greatly subject to human judgement and human error. The data fed into the system is done so by humans, and it is humans who decide to deploy technology despite these low accuracy rates. Similarly, how data is classified, the technically embedded rules used to process it, and the desired outputs are all human-driven factors.

In the context of Israeli attacks on Gaza and Lebanon, one of the key questions to ask with relation to civilian death and displacement, as well as with the destruction of civilian infrastructure, is precisely one of data classification. Both Hezbollah and Hamas are deeply implicated in civilian life through such actions as education, social protection, money-lending, and politics, while simultaneously being non-state paramilitary groups. Nonetheless, Hezbollah – in all of its iterations – has been designated as a terrorist group by the US and several European countries. This raises a key question: are non-armed, civilian affiliates or infrastructure of these groups, such as those who distribute food or financial support to community members, included in lists generated by AI-enabled systems?

While we do not have a clear answer, these points are important as we look at the supposed “mistakes” of AI-enabled systems in warfare; while there are certainly technological errors and clear problems with accuracy rates, many targets and classifications are defined by humans and in pursuit of human goals.

5)  What does "human oversight" mean in these contexts? Aren't humans supposed to double-check AI-generated lists in order to ensure that no civilians are included?

We do not know to what extent humans are involved in the AI-enabled processes used in the US or Israeli militaries. Current reporting would suggest that it depends on the system, with humans being involved both in the confirmation of targets and the overall approval for systems to be used. However, given the extremely high number of targets and strikes in the current war, it is highly likely that any human involvement is being greatly impacted by a phenomenon known as automation bias.

Simply put, automation bias is the well-documented "tendency to over-rely on automated recommendations". While relevant in all fields that undergo automation, in the Israel-US-Iran war this could include target lists produced by programs such as MSS or The Gospel. This occurs because humans often perceive algorithms as being based on objectively “true” data and therefore as producing optimal analyses and suggestions. While humans may be intervening by technically providing their approval for kill lists, automation bias suggests that there is little scrutiny of names presented or the underlying data leading to these suggestions. In this sense, human presence is often used to “rubber stamp” without actually providing useful supervision and balances. As the answer to Question 4 shows, this lack of attention can be extremely dangerous and often leads to widescale civilian death.

Similarly, the mere presence of human oversight does not immediately mean that there is “humanity” in decision making. While humans may be officially within a certain decision-making process, the presence of biased data inputs and automation bias for outputs means that mechanisms to ensure the making of ethical decisions are weak.

In light of the above, the question is not necessarily one of how human intervention is defined in theory but rather the degree to which militaries such as the US and Israel are delegating their decision making to effectively non-human processes in practice.

6)  Are there no legal barriers to the use of AI in warfare?

There is an ongoing and polarized debate regarding the capacity of existing legal systems to control AI and technology in warfare. Fundamentally, this debate questions whether new treaties are indeed necessary or if existing laws can be used to regulate and apply guardrails to the use of AI in contexts of conflict.

As this debate unfolds, in a world where International Humanitarian Law has been heavily criticized – particularly following 7 October 2023 – and calls for accountability by international legal institutions are routinely ignored, accountability and oversight has shifted to national arenas. Several states such as Australia, the US, and the UK have adopted national frameworks that seek to define what is considered acceptable use for AI by their military complexes.

These legal developments, however, exist within a larger context where human factors continue to impede meaningful accountability for crimes committed by the military. The main obstacle to accountability, therefore, is often not the absence of rules but rather the weakness of the institutions and political will needed to enforce them. In practice, national AI frameworks for military use tend to function more as statements of principle than as binding and effective constraints.

Even if there were to be political will on behalf of the military to hold its members to account, the questions remain around through which processes and for which acts. Australia's "Policy Settings for Responsible Use of Artificial Intelligence in Defence", for example, provides a high-level framework but with little emphasis on implementation mechanisms, similar to its US counterpart of 2020. This lack of clarity makes prosecuting more difficult and impedes eventual accountability processes.

Compounding this is a high level of opacity relating to the specific acts or people responsible for a certain strike, with experts noting that the technical processes behind a given system are often too complex and not understood well enough to allow for clear attribution of blame.

As such, while there may be legal frameworks and international conventions that push for lawful deployment and use of AI-enabled weaponry in contexts of war, the challenge is at its heart not only one of legal constraints. There are three human hurdles that impede effective accountability: clarity on implementation, delineated responsibility, and the willingness (on behalf of the military complex) to hold military members accountable. As Nehal Bhuta, Professor of International Law at the University of Edinburgh, has previously stated: "We're going to get as much lawfulness as we commit ourselves to in these processes".

7)  Does the recent Anthropic/Pentagon feud have anything to do with current events?

The dispute between Anthropic and the Pentagon is key to the US-Israel-Iran war precisely because Claude AI is integrated in MSS and, according to all relevant accounts, is being used to accelerate target nomination and selection.

The fracture began in January 2026, where reports suggested that Claude was used for real-time data processing during the raid to capture Venezuelan President Nicolás Maduro. After a reported heated back-and-forth between Anthropic executives, who rejected the use of Claude AI for mass domestic surveillance and full autonomous weaponry, and the US Department of Defense (DoD), Defense Secretary Pete Hegseth ultimately ended up classifying Anthropic "a supply chain risk". Concretely, this would mean that the DoD must "phase out their use of [Anthropic's] products" over a six-month period. Within hours, OpenAI signed a deal to replace Anthropic, reportedly including the very same restrictions the Pentagon had just punished Anthropic for requesting.

Changing providers, however, does not change the issues embedded in these systems. The dissociation of Claude AI and the integration of Open AI are both technological processes that may also produce errors and cause harm to civilians on the ground. In short, while this debate is happening on a nuts-and-bolts level, it is highly unlikely that the end result will result in more accurate targeting or reduced civilian harm.

8)  What is the impact of this war on the infrastructure that supports AI?

AI and cognitive computation have become increasingly coupled with critical infrastructure and essential services, especially in the GCC as countries double down on the implementation of their digitalization strategies. The most prominent example is data centers. While generally classified as civilian infrastructure, recent attacks have caused debate around whether or not data centers can be considered as "war infrastructure". However they are defined, this physical presence is vulnerable and can be disrupted by low-cost drone activity. When data centers are disrupted, every sector that utilizes these services is impacted; in the GCC, this is especially true of the chemical sector, with high levels of digitization and AI adoption.

Recent strikes on AWS data centers in the UAE and Bahrain illustrate this shift. When unidentified objects struck an AWS facility in the UAE, the resulting fire triggered outages that cascaded through the financial sector. Abu Dhabi Commercial Bank, for example, saw its mobile apps and platforms go offline. Similarly, at least one banking data center in Iran has been struck, potentially resulting in service outages.

The infrastructure behind AI, however, is not limited to data centers: connectivity, energy, compute capacity, economic power, private-public partnerships, and a specialized workforce, in addition to data, all form an "Architecture of AI". Despite the GCC's focus on these pillars to cement its place as a future leader in the field and its access to large amounts of energy, regional instability is presenting clear issues for connectivity and a specialized workforce.

The problem extends beyond data centers to the 17 submarine cables passing through the Red Sea, which carry the majority of data traffic between Europe, Asia, and Africa. With the closure of the Strait of Hormuz and renewed Houthi threats in the Red Sea, both critical data chokepoints are simultaneously in active conflict zones. This geographic reality means that even if a data center remains operational, its supporting environment can be massively disrupted at any moment.

Similarly, Ragheb Ghandour, CyberSecurity Technologist at SMEX, explained that submarine cables are "critical for data center operability," yet the Red Sea remains a critical geography where disruptions expose a huge dependency on the UAE’s role as a “middle infrastructure” for Europe-Asia connectivity. This systemic risk was further mapped by Gilbert Traverse, Managing Director at Oracle in the Office of Technology and Innovation, who identified a dangerous overlap between nine active conflict zones and the world’s primary data center hubs and cable chokepoints. His analysis warns that while the industry successfully "optimized global compute for latency and cost", it failed to build an architecture resilient enough for a "world at war".

In addition to the vulnerability of physical infrastructure, the war has also exposed fragilities in the foundational inputs needed to build and sustain it. The current situation has brought to the fore an imminent helium shortage, which is essential for maintaining optimal conditions during various stages of chip manufacturing, with global implications. It has also highlighted attacks on water infrastructure such as desalination plants, which are key elements in cooling processes for data centers.

These examples demonstrate how the impact of war extends beyond immediate and local impact to having global ramifications on connectivity and the production of future hardware. In addition to the vulnerability of these globally important supply chains, a key takeaway is the increasingly dual-use nature of certain infrastructures and potential implications this could have in future conflicts.

9)  What else has been happening in the MENA tech ecosystem since the start of the war?

While the current war is taking up a large part of media attention, the tech world has not stopped turning in the region. As an inexhaustive list, Oman has talked about building its own national AI supercomputer, Algeria's telecommunications minister attended a high-level meeting on AI governance in Barcelona, and Mauritania has said that its new subsea cable will most likely be completed by 2027. Despite the multiple strikes on data centers in the region, private company Infobip has opened a data center in Riyadh, Microsoft has confirmed the expansion of its services in Saudi Arabia, and Egypt's Renergy Group could establish a data center in the Sinai Peninsula, though details about the development are few and far between. As of 2 March 2026, there was a large drop in AI and tech-related stocks that are based in the region. Regardless, Visa has created a sub-regional strategic hub covering Libya, Egypt, and Sudan to strengthen its presence in North Africa and Starlink has expanded to Kuwait and the UAE.

Moving forward, ARI will be keeping close watch of how the increasing layers of private-public investment and development shape the ecosystem, with data becoming increasingly monetized but less private.