BREAKING NEWS | WORLD DESK | US-IRAN WAR / AI & NATIONAL SECURITY.
, Anthropic Warns In New Threat Report
A 154-page threat intelligence report from the AI company documents an Iran-linked operation that mined public data through Claude to track US naval assets in the Middle East, part of a broader accounting of nation-state misuse that also implicates China, Russia and Yemen’s Houthi movement.
WASHINGTON — The Atlantic Digest
Iran-linked operators used Claude, the American-developed artificial intelligence model built by Anthropic, in an attempt to compile targeting profiles on United States Navy vessels operating in the Middle East, according to a 154-page threat intelligence report the AI company published on Thursday, warning of what it called emerging risks to national security from the misuse of frontier AI systems.
The report, titled “Detecting and countering misuse of AI: September 2026,” covers threat activity Anthropic says it identified and disrupted between roughly December 2025 and August 2026. Its most striking finding describes an Iran-nexus actor that used Claude to build what amounted to a targeting handbook on US naval assets: compiling rosters of American personnel scraped from captions on publicly posted military photographs, collecting ship and aircraft transponder identifiers from open sources, running commercial satellite-imagery queries, cataloguing websites that expose naval movements, and researching vulnerabilities in shipboard communications systems, including VSAT satellite terminals, Cisco networking equipment and industrial control systems.
“Without the correct safeguards, such capabilities could have catastrophic consequences,” Anthropic said in the report. The company said it identified the account responsible, banned it, strengthened its safeguards based on what it learned from the case, and shared intelligence on the operation with government authorities. Anthropic was careful to note a limit to what it could establish: the company said it could not confirm whether the open-source intelligence the operation compiled was ever put to use in a real-world military or operational context.
The Navy-targeting case was one of several nation-state and criminal operations detailed in the report. Anthropic said Chinese-speaking operators, tracked under the designation GTG-10007, conducted sustained research against a major security product and produced multiple previously unknown vulnerabilities; separate press reporting on the document has described a China-linked actor separately using Claude to help prepare briefing material on US anti-torpedo weapons systems, though Anthropic’s own published account, as reviewed by The Atlantic Digest, did not independently confirm that specific detail. The report also documented a Russian-linked espionage operation, GTG-20006, that used “AI-driven workflows that automated much of their operations” against military intelligence targets in Ukrainian and European governments.
On Iran specifically, the report went well beyond the Navy-targeting case. Anthropic said a separate Iran-nexus actor built an automated, open-source intelligence identity-profiling system aimed at hundreds of Israeli governmental and non-governmental individuals as well as Jewish diaspora organisations, while two further Iranian-linked actors used Claude to surveil and profile more than 6,000 Iranian citizens, running social-network analysis across more than 155,000 social media posts to build profiles of minorities, dissidents and regime opponents. The report also described an influence-operations campaign in which an Iranian state-linked body, identified as the Islamic Culture and Communications Organisation, used AI to help build operational manuals, generate personas and convert government material into targeted propaganda distributed under a campaign Anthropic referred to by the Arabic phrase “Jihad al-Tabyin.”
Elsewhere in the report, Anthropic said a Yemen-based group it linked to Houthi networks attempted to use Claude for research supporting ballistic-missile guidance software, and that its safeguards blocked many, though not all, of the group’s requests. The document also catalogued financially motivated misuse, including an extortion and data-theft operation Anthropic tracked as GTG-50014 and linked to the ShinyHunters criminal group, alongside AI-assisted scam operations and attempts to steal proprietary AI capabilities.
Anthropic framed the broader pattern as a shift in the economics of state-sponsored and criminal activity rather than the emergence of fundamentally new attack techniques, saying AI “has collapsed the labor and tooling gap” that previously separated well-resourced state operations from individual actors. The company described the Navy-related case as among “the most notable and novel threat activity we’ve identified to date,” while cautioning that “sophisticated and persistent threat actors continuously test our safeguards.”
The disclosure lands amid heightened US military vigilance in the Middle East. Acting Navy Secretary Hung Cao issued guidance to personnel on August 19 directing sailors to strip Navy affiliation details from social media accounts and to report drone sightings near installations, suspicious surveillance activity, and any attempts by outsiders to obtain information about ships, personnel or deployments. “By embracing this heightened posture of vigilance, we will safeguard our force, protect our families,” Cao’s message said, referencing concerns over “uncrewed aircraft systems encounters, suspicious activity, cyber threats and acts of violence.”
The report’s publication comes roughly six months into the Strait of Hormuz crisis, which began on February 28 when the United States and Israel launched coordinated strikes on Iranian military and nuclear sites under Operation Epic Fury, prompting Iran to move to block the Strait of Hormuz, a chokepoint that normally carries about a fifth of the world’s oil and liquefied natural gas trade. In the six months since, more than 65 merchant vessels have been attacked or seized amid the standoff, and US naval forces have conducted blockade and escort operations against Iranian naval and paramilitary forces in the Gulf, a backdrop that gives added weight to Anthropic’s account of an Iran-linked effort to map US Navy positions and vulnerabilities using commercially available AI tools.
Anthropic said it continues to work with US government authorities and industry partners on detecting and countering this kind of misuse, though the report acknowledges the limits of any single company’s enforcement: accounts can be created under false pretences, and access to AI models can be routed through intermediaries in ways that complicate detection. The Atlantic Digest was unable to reach the Pentagon, the US Navy, or Anthropic for further comment beyond the material contained in the published report at the time of this story, and will update this report as more detail becomes available.
Sources
- Anthropic — “Detecting and Countering Misuse of AI: September 2026” (threat intelligence report)
- Stars and Stripes — “Iran-Linked Actor Used AI To Build Targeting Guides On US Navy”
- The Jerusalem Post — “Iran Used AI To Develop Military, Propaganda Strategies, Anthropic Report Reveals”
- Crypto Briefing — “Anthropic Warns Iran Used US AI Models To Target US Naval Ships”
- The Epoch Times — “Iran Used Claude AI Model To Target US Navy Warships: Anthropic Report”
- Yahoo News — “Iran Used Claude To Target US Navy In Middle East, Anthropic Says”
- Wikipedia — “2026 Strait of Hormuz Crisis”
This report is developing. Findings described above are as published by Anthropic in its own threat intelligence report; Anthropic states it could not confirm whether the compiled intelligence was used operationally, and one detail regarding Chinese-linked anti-torpedo research is attributed to secondary press reporting rather than Anthropic’s own published text.
