The revelation that Iranian state-linked threat actors utilized Anthropic’s flagship AI model, Claude, to build targeting packages against US naval forces, map Israeli institutions, and code guidance software for Houthi ballistic missiles marks a watershed moment in modern warfare. What was once dismissed as a hypothetical threat scenario by Silicon Valley ethicists is now an active operational reality in the Middle East. The 154-page threat assessment published by Anthropic provides damning, undeniable proof that commercial artificial intelligence developed in the West has been weaponized directly by the Iranian regime and its regional proxies.
For the defense establishment in Jerusalem, the implications of this report are immediate, profound, and deeply alarming. The primary danger of generative AI in the hands of an ideological adversary is not merely that it creates clever propaganda or automates phishing emails. The true threat lies in its ability to act as a massive force multiplier for hostile intelligence gathering. Iranian military operators used agentic AI workflows to conduct complex multi-victim cyber campaigns that previously required entire teams of human intelligence analysts, systems engineers, and regional specialists.
The scope of Tehran’s intelligence gathering uncovered by Anthropic reveals a highly systematic, industrial-scale operational pipeline. Iranian operatives utilized AI models to scrape public military photograph captions, compile rosters of active United States service personnel, parse public transponder data, and execute commercial satellite imagery queries. They simultaneously probed vulnerabilities in critical shipboard hardware, including maritime VSAT satellite terminals, Cisco routing gear, and industrial control systems operating across naval platforms. The machine learning pipeline transformed raw, disconnected open-source data into refined, actionable military targeting lists at speed and scale.
Equally troubling for Israeli defense planners is how Tehran deployed these automated intelligence tools against the Jewish state and the global diaspora. Iranian network actors created automated open-source profiling systems that targeted hundreds of Israeli government officials, civil society leaders, non-governmental figures, and Jewish diaspora organizations worldwide. By leveraging agentic AI to analyze social connections and harvest personal data, Tehran built an automated intelligence factory. This framework allowed operatives to write tailored, highly evasive malware and launch precision intelligence operations that bypass traditional cyber defense perimeters.
This technological leap extends well beyond intelligence collection into the physical realm of advanced weaponry. Anthropic confirmed that a Northern Yemeni faction, the Iranian-backed Houthi movement currently holding a stranglehold on international shipping in the Red Sea, used Claude to assist in developing guidance software for high-end military hardware, including multi-stage ballistic missiles. While internal safeguards blocked several query attempts, the company admitted that safety guardrails failed to stop every request. The uncomfortable truth is now exposed to the world: Western commercial software has actively contributed to refining the accuracy of missiles fired at international shipping lanes and Israeli territory.
At home, the regime in Tehran has integrated these advanced systems into its apparatus of domestic oppression and cognitive warfare. Within a single year, Iranian actors utilized the model to track and profile 6,388 Iranian citizens. They constructed dedicated social network analysis pipelines to process over 155,216 social media posts, combining automated license plate recognition, interception of mobile device identifiers, and scrape files from private Telegram channels to suppress political dissent. This internal surveillance infrastructure was paired with “Jihad al-Tabyin,” or explanatory jihad, an automated soft-war strategy executed by organizations like the Islamic Culture and Communications Organization. Their goal was clear: churn out IRGC-aligned propaganda manuals and weaponize online messaging campaigns across foreign democracies.
This intelligence crisis highlights a fatal structural flaw in how Silicon Valley manages dual-use technologies. While leading tech firms market their models as general-purpose productivity utilities, they are, in functional reality, advanced military systems capable of automating complex cyber operations. The internal dissent at Anthropic, highlighted by the recent resignation of senior researcher Evan Hubinger after he publicly warned that AI risks remain unresolved, illustrates a corporate sector struggling to maintain control over its own creations. Silicon Valley companies cannot simultaneously decline cooperation with Western defense agencies over domestic surveillance concerns while inadvertently providing military capabilities to anti-Western dictatorships.
From a strategic perspective, the traditional paradigm of export controls, air-gapped systems, and software bans is failing. Foreign intelligence services do not need to steal source code or breach secure facilities when they can simply access commercial API endpoints, deploy automated prompt engineering techniques, and bypass content filters through external proxies. When a rogue actor can query an American server to optimize missile guidance parameters or map military personnel rosters, the distinction between civilian software and military hardware completely collapses.
Jerusalem and its Western allies must rapidly adjust their defense posture to meet this new reality. Israel’s defense apparatus can no longer evaluate regional threats strictly by counting missile stockpiles, drone production facilities, or enrichment centrifuges in Iranian facilities. The technological sophistication of the threat is now dictated by the computational power, data access, and model capabilities available to adversary forces. Countering this threat requires real-time coordination between intelligence agencies and major software developers to identify, isolate, and terminate malicious user pipelines the moment unusual operational patterns emerge.
The Anthropic report serves as an urgent wake-up call to the global intelligence community. The democratization of artificial intelligence has effectively erased the technological gap that long protected Western militaries from asymmetrical adversaries. Tehran has demonstrated that it possesses both the technical sophistication and the strategic intent to integrate artificial intelligence directly into its war machine. If Israel and its allies fail to establish aggressive counter-AI defense protocols and hold software developers strictly accountable for their supply chains, the tactical balance in the Middle East will continue to shift dangerously toward those who seek its destruction.