When the war in Ukraine broke out in 2022, it was generally believed that this was the first conflict to fully utilize modern technology on the battlefield; and the current war against Iran is regarded as the first war in which artificial intelligence plays a central role, including participating in the planning and execution of air strikes at a pace faster than "human thinking speed."
Multiple reports this week stated that the United States and Israel used Anthropic's Claude large model in their early joint operations against Iran for intelligence analysis and target-related scenario deductions and action plan planning. This news once again triggered concerns from the outside world: large language models are rapidly being included in the so-called "kill chain" (kill chain), and the decision-making chain is greatly accelerated by machines. Human commanders are facing huge pressure on whether to adopt AI options before the traditional supervision procedures are completed.
Reports indicate that Claude was used to assist in the first round of strikes against Iran on Saturday, which struck multiple targets and resulted in the death of Iran's Supreme Leader Ayatollah Ali Khamenei. Meanwhile, the U.S. military said it was investigating reports in Iranian state media that a missile hit a school in the south, killing 165 people, many of them children.
This comes just days after the Trump administration designated Anthropic as a "supply chain risk" company, requiring federal agencies and the military to stop using the company's tools. Previously, negotiations between the two sides broke down on the boundary issue of AI use: Anthropic insisted that it "not be used for large-scale domestic surveillance of U.S. citizens" and "not be used for fully autonomous weapons systems," while the U.S. demanded that the defense sector must have "full, unrestricted access" to related technologies.
Although the Trump administration has ordered a "supply cutoff," Anthropic's tools continue to be used in some military systems, but related deployments are gradually being replaced by OpenAI's models. It is reported that OpenAI reached a new cooperation agreement with the Pentagon this past weekend to provide alternative AI capabilities for the US defense system.
As early as 2024, Claude was integrated into a system developed by the war technology company Palantir for the U.S. Department of War and other national security agencies. The goal is to "significantly improve the efficiency of intelligence analysis and provide support for officials' decision-making." This means that AI has been upgraded from an auxiliary tool to an indispensable part of the process in intelligence collection, pattern recognition and action suggestions.
Craig Jones, a senior lecturer in political geography at the University of Newcastle in the United Kingdom and an expert on "kill chain", pointed out in an interview with the Guardian that AI can already give suggestions on targets. "In some aspects, its speed is actually faster than human thinking." He emphasized that in this kind of system, the scale and speed of war are amplified: while carrying out "targeted killing" strikes, it also weakens the opponent's ability to retaliate through air and ballistic missiles at almost the same time. Such operations that would take days or even weeks to complete in traditional wars can now be "completed simultaneously."
At the same time, Iran claimed in 2025 that it used independently developed artificial intelligence technology in its missile guidance and target positioning systems. However, judging from current public signs, Iran’s main investment in AI is still focused on cyber warfare: including using AI to generate phishing emails, launching distributed denial-of-service attacks (DDoS), and carrying out other types of destructive network intrusions against U.S. targets, while using algorithms to amplify the effects of propaganda and public opinion warfare.
Overall, the role of artificial intelligence in modern warfare is no longer a "supporting role", but is becoming a core component of the offensive and defensive system, significantly compressing the time window between reconnaissance and surveillance, intelligence analysis and implementation of strikes. In addition to the real risk that AI may still go "seriously wrong" in complex environments, the outside world is more worried about how this usage model will further upgrade in the future, and the extent to which humans can control the power they unleash in the absence of adequate international norms and checks and balances.