We, the undersigned organizations and individuals, are deeply alarmed by the rapid militarization of artificial intelligence (AI) technologies.
AI systems embedded into military kill chains are accelerating the speed and scale of military assaults in a manner that creates significant new risks for accountability in conflict and risks facilitating violations of international criminal, human rights, and humanitarian law. We therefore call for tech companies and states to halt the provision of AI systems for use in the military kill chain and to take all steps to ensure that other AI systems they provide do not cause or contribute to violations of international humanitarian law (IHL) and international human rights law (IHRL). This includes the use of AI decision-support systems, including target generation systems, remote biometric surveillance, and multimodal AI models, including large language models (LLMs). AI-accelerated warfare is rapidly becoming a means of rubber-stamping killing at speed and at scale, and currently no technical or procedural fixes can effectively prevent the lethal and devastating consequences that stem from the fundamental challenges it poses to international law.
All companies, including those contracting with government military agencies across the AI supply chain, from licensing and training ‘frontier’ models to providing data processing and storage functionalities, must take all possible steps to ensure that their products and services are not causing, contributing to, or being directly linked to human rights abuses and international crimes. In armed conflict, this responsibility extends to respecting international humanitarian and criminal law, given the heightened risk of facilitating gross human rights abuses, including grave violations of international law, in such contexts. Where companies cannot meaningfully prevent or mitigate such risks, they must not enter into or fulfill such contracts.
AI-enabled data storage and analysis systems used in the kill chain, including Anthropic’s Claude large language model and the Maven Smart System, are, according to an NBC investigation, playing a role in supporting U.S. and Israeli attacks on Iran. Open AI recently agreed to provide AI services to the U.S. Department of Defense (DoD); Google has contracted with the DoD, like Anthropic, to “develop prototype frontier AI capabilities to address critical national security challenges in across warfighting and enterprise domains”; Microsoft, Google, and Amazon have for years provided data storage, processing, and other enterprise infrastructure services to DoD “warfighting” programs.
According to media reports and official statements by the DoD, the rapid generation of targets by AI tools has enabled the increase in speed, scale, intensity, and destructive force of U.S. strikes on Iran. Within the first 48 hours of strikes, Israel and the U.S. reportedly struck nearly 2,000 targets in Iran. While much remains unclear about the precise role played by AI systems in these military attacks on Iran, the strikes have had a devastating impact on civilians and civilian infrastructure.
The adoption of AI targeting systems in this campaign follows the example of the Israeli government’s weaponization of data analysis and machine learning tools, powered by mass surveillance, in its genocidal attacks on Gaza. By diluting human responsibility for life-and-death decisions, Israel’s use of systems such as Lavender, Gospel, and Where’s Daddy may contribute to the obfuscation of international crimes behind a veneer of perceived algorithmic objectivity while also obfuscating accountability.
Not for the first time, we are seeing Palestine used as a laboratory for experimental and dehumanizing methods of warfare, including through corporate tech partnerships with Israeli military agencies. Microsoft, Google, Palantir, and other tech companies may have contributed to or enabled the Israeli government’s access to mass data storage, processing and analysis systems that are aiding their ongoing destruction and genocide in Gaza, which has so far led to the killing of at least 72,000 Palestinians.
Legal scholars and practitioners, technical experts, tech workers, UN special rapporteurs, and investigative journalists have long warned against the development and deployment of AI in warfare, given the heightened risk of international crimes. Despite claims by their proponents that AI tools are making warfare more effective, precise, or humane, real-world deployments indicate that AI is actually facilitating more violent, dehumanizing, and destructive methods of warfare.
In particular, we are profoundly concerned that the use of LLMs for target generation and prioritization is pushing military actors into a form of warfare where foundational principles of international humanitarian law—including the principles of distinction, proportionality and precaution—are not, and arguably cannot be, sufficiently respected, given the sheer speed and scale of such technologies, in addition to the unreliable, biased, and often illegally obtained input data. And we further assert that these dynamics risk facilitating human rights abuses, crimes against humanity, and war crimes. Moreover, the opacity that comes with the use of these tools fundamentally threatens the possibility of attributing moral or legal responsibility in cases where errors are made. As Anthropic has stated themselves, “…today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk.” Actors who choose to deploy AI systems that are used to commit international crimes must be held criminally responsible.
Our concerns are not limited to the errors that may result from such systems malfunctioning but encompass how these systems fundamentally transform military operations. We therefore reject the premise that at present, technical or functional fixes, whether that be a purported ‘human in the loop’ or supposedly hardcoded guardrails in AI models, can prevent the lethal and devastating consequences of AI-accelerated kill chains. These propositions allow for the normalization and proliferation of embedding AI in military decision-making to the detriment of vulnerable communities and populations. At present, meaningful human control, genuine accountability, oversight, and transparency of these technologies are not possible in their current form.
Even where AI systems used for target acquisition do not make the final decision to kill, they risk becoming rubber-stamping mechanisms for killing at scale because they appeal to false notions of objectivity and may displace accountability and due diligence, which can ultimately serve to expedite and streamline mass killing. Layering such systems with even more ‘frictionless’ techniques of surveillance, target acquisition, and command operations, e.g., in the form of large AI models, such as LLMs, automates dehumanization by reducing questions of life and death to a simple chat prompt. The decision to kill another human being carries a grave moral and legal weight and must never be devolved to purely accepting or rejecting the recommendations of AI systems. When militaries rely on AI to expedite target identification with such speed and routinization that any human review risks becoming a rubber stamp without meaningful human control, mass killing can and will often ensue, in direct violation of the principle of precaution under IHL.
Companies have a responsibility to respect human rights and refrain from causing or contributing to human abuses and other violations of international law, including providing material or financial support to states engaged in international crimes. As reflected in the UN Guiding Principles on Business and Human Rights, companies engaged in such conduct must immediately cease their contribution to harm. Even when a company is not causing or contributing to harm but is merely linked to it, it is expected to use its leverage to seek to bring an end to these violations.
Once they have entered military contracts, companies may have limited agency over how their products and services are used, demonstrated through Anthropic’s standoff with the US government, as well as reports of Google and Amazon suspending the applicability of their terms of services in contracts with the Israeli government. As recently as April 27, 2026, more than 560 Google employees signed an open letter to Google’s CEO, urging the company to refuse to let the US government use its AI technology in classified military operations.
Tech companies and their executives should take seriously their potential culpability in cases where their technologies play a role in violations of international law before entering these lucrative defense contracts, and refrain from doing so where they cannot make such an assessment. Beyond that, they must also understand the role they have to play in reshaping the normative architecture of the use of AI in conflict.
We, the undersigned organizations and individuals, call for:
Tech companies to:
- Refrain from entering into or fulfilling contracts with military agencies or armed groups that commit possible violations of international law, including human rights violations and atrocity crimes;
States to:
- Refrain from selling, transferring, servicing, or exporting AI decision support systems for military kill chains and human targeting, including target generation systems and remote biometric surveillance; and
- Refrain from selling or exporting AI decision-support systems for non-lethal purposes, including multimodal AI models such as LLMs, for use in military decision-making processes, until genuine accountability, meaningful human control, oversight, and transparency is made possible in line with principles of international humanitarian law and international human rights law.
- Halt the use of AI tools, including large language models, in the conduct of military targeting, and ensure adherence to principles of international humanitarian law and international human rights law; and
- Provide transparency on how AI is currently being used in the conduct of hostilities.
The full list of signatories is available on Access Now: https://www.accessnow.org/press-release/joint-statement-on-ai-in-warfare/
Amnesty International wrote to Open AI and Anthropic relating to their human rights policies and practices in relation to generative AI use in military contexts, as it has not previously corresponded with these companies on this topic. At the time of publication of this statement, only Open AI responded. Their response is available here.
Read the joint statement.