How AI Like Claude Is Being Used by Iran and the Houthis

Breaking: Houthi Rebels Used Claude AI to Build Guided Missiles — Anthropic Confirms

In a revelation that has sent shockwaves through the global technology and defense communities, Anthropic — the creator of the widely-used AI assistant Claude — confirmed that users in Houthi-controlled northern Yemen attempted to use its AI model to develop advanced guided missiles and ballistic weapons.

The disclosure, made in Anthropic's September 2026 Threat Intelligence Report, marks one of the most alarming cases of AI misuse to date and raises urgent questions about the weaponization of civilian AI systems by state-aligned non-state actors.

Key Finding: A weapons engineering cell in northern Yemen simultaneously pursued three weapons programs using Claude AI, including guided rocket development that progressed to a failed live test.

What Actually Happened: The Full Timeline

The Discovery

Anthropic's threat intelligence team identified a coordinated operation in northern Yemen — territory controlled by the Iran-backed Houthi movement (Ansar Allah) — where multiple Claude accounts were being used for weapons-related purposes.

The three concurrent weapons programs identified:

1. Guided Rocket Development — Software for missile guidance systems
2. Ballistic Missile Guidance — Advanced trajectory calculation algorithms
3. Weapons Testing Protocols — Test procedures and failure analysis

The Live Test

The operation went beyond theoretical research. According to Anthropic's report:

• The Yemen-based cell conducted a live test of a guided rocket
• The test failed — the weapon did not achieve its intended trajectory or target
• Within hours of the failed test, the operators returned to Claude to investigate what went wrong
• They used the AI to analyze the failure and attempt to diagnose the problem

Anthropic's Response

Upon identifying the accounts, Anthropic immediately blocked them. The company stressed that:

• The users did not succeed in "fielding an operational device"
• There was no evidence of a successful weapons deployment
• The accounts were terminated and reported to relevant authorities

The Iran Connection: How Tehran Enables Houthi AI Warfare

Iran's Technology Pipeline to the Houthis

The Houthi movement's access to AI tools doesn't exist in isolation. It's part of a broader Iranian technology transfer pipeline that has been documented extensively by the United Nations, U.S. intelligence agencies, and independent research organizations.

Iran's support for the Houthis includes:

The AI Dimension

The use of Claude AI represents a new frontier in Iran's support for the Houthis. Rather than shipping physical weapons components — which are subject to interdiction by the U.S. Navy and coalition forces — AI provides:

• Remote capability delivery — No physical shipment required
• Deniability — AI usage is harder to trace than weapons shipments
• Scalability — A single AI model can be accessed by unlimited users
• Adaptability — AI can be repurposed for different weapons programs

China's Role: The Silent Partner

According to the Soufan Center and U.S. Treasury Department sanctions, China has also become a significant source of technology for Houthi weapons production:

• Guangzhou Tasneem Trading Company — Procured materials for weapons manufacturing
• Ningbo Beilun Saige Machine Co. — Coordinated drone component procurement
• Dongguan Yuze Machining Tools — Supplied equipment for domestic weapons production

China has not been identified as directly advising or arming the Houthis, but Beijing has turned a blind eye to procurement activities within its borders.

The U.S. Military's Own AI Dilemma: Claude in the Iran War

The Pentagon's Secret Usage

While the Houthis were using Claude to build weapons, the U.S. military was simultaneously using the same AI model for its own operations against Iran.

Timeline of events:

• February 28, 2026 — U.S.-Israel strikes on Iran begin; Claude is reportedly used for intelligence analysis, target selection, and battlefield simulations
• March 1, 2026 — President Trump orders all federal agencies to stop using Anthropic's technology
• March 3, 2026 — CBS News confirms the military continued using Claude despite the ban
• March 4, 2026 — Pentagon declares Anthropic a "supply chain risk"

The Anthropic-Pentagon Conflict

The dispute centered on guardrails that Anthropic sought to impose:

Anthropic's Position:

• Claude should not be used for mass surveillance of Americans
• Claude should not power fully autonomous weapons
• A human must make the final decision in lethal operations

The Pentagon's Position:

• Demand for unrestricted access to all AI models for "all lawful purposes"
• It's already illegal for the Pentagon to conduct mass surveillance of Americans
• Internal policies already restrict fully autonomous weapons
• "At some level, you have to trust your military to do the right thing" — Pentagon CTO Emil Michael

The Minab School Strike

The most controversial incident involved the February 28 missile strike on Shajareh Tayyebeh Elementary School in Minab, Iran, which killed an estimated 120 children.

In a Bloomberg interview, Anthropic CEO Dario Amodei admitted:

"We don't have access to, we don't know exactly how these models were used."

He described the strike as "a really terrible thing to happen" but maintained the use case did not violate Anthropic's policies, citing the principle that "a human makes the final decision."

The paradox: If Anthropic doesn't know how Claude was used, it cannot verify whether the human-in-the-loop principle was actually followed.

The Maven Smart System: Where AI Meets Lethal Force

Palantir's Targeting Platform

Claude is embedded in the Maven Smart System, an AI-assisted targeting platform built by Palantir Technologies under a $1.3 billion Pentagon contract.

System capabilities:

• Process intelligence data from multiple sources
• Identify and prioritize potential targets
• Run battlefield simulations
• Support real-time decision-making

Operational statistics:

• 1,000 targets struck in the first 24 hours
• 13,000 targets reached by April 6, 2026

The Visibility Gap

The Maven Smart System exposes a critical structural problem in AI governance:

1. Anthropic licenses Claude to defense contractors
2. Palantir deploys Claude in classified military systems
3. Anthropic has no visibility into how the model is used
4. Published safety principles become unverifiable commitments

This gap isn't unique to Anthropic — it's the fundamental challenge facing any AI developer that licenses foundation models to defense contractors.

The Broader Crisis: AI in Modern Warfare

The Global AI Arms Race

The Houthi case is not an isolated incident. AI is transforming warfare across multiple theaters:

Ukraine-Russia Conflict:

• AI-powered drone targeting systems
• Automated surveillance and reconnaissance
• Predictive maintenance for military equipment

Israel-Gaza War:

• IDF's "Lavender" AI targeting system
• Automated target identification and tracking
• AI-assisted intelligence analysis

U.S. Operations:

• Maven Smart System for targeting
• AI-enhanced logistics and supply chain management
• Intelligence synthesis and analysis

The Democratization of Lethal AI

What makes the Houthi case particularly alarming is the accessibility of the technology:

Traditional weapons development:

• Requires physical materials and facilities
• Subject to export controls and interdiction
• Needs specialized engineering expertise
• Takes years to develop

AI-assisted weapons development:

• Requires only internet access and a computer
• No physical materials to intercept
• AI provides the engineering expertise
• Can compress development timelines dramatically

Ethical Implications: The Tech Industry's Reckoning

Anthropic's Ethical Framework

Anthropic has positioned itself as a safety-first AI company, emphasizing:

• Constitutional AI — Built-in ethical guidelines
• Responsible Scaling Policy — Graduated deployment based on capability
• Red Lines — Explicit prohibitions on harmful use cases

The company's stated prohibitions:

• Using Claude for violent ends
• Developing weapons
• Conducting surveillance
• Enabling human rights abuses

The Enforcement Problem

The Houthi case exposes the fundamental limitation of AI ethics frameworks:

1. Account verification — How do you verify who is using the AI?
2. Intent detection — How do you distinguish between research and weapons development?
3. Usage monitoring — How do you monitor conversations without violating privacy?
4. Enforcement — What happens when you detect misuse?

Anthropic's response — blocking accounts after detection — is reactive rather than preventive. The weapons development had already occurred before the accounts were identified.

The Industry Response

The revelation has sparked debate across the AI industry:

OpenAI's Position:

• CEO Sam Altman reached agreement with the Pentagon for classified network use
• Emphasized partnership with defense sector
• Less vocal on ethical restrictions than Anthropic

Google's Position:

• Employees pushing for guardrails around military AI use
• Internal debate about defense contracts
• Historical reluctance after Project Maven controversy

Meta's Position:

• Open-source approach raises different challenges
• Llama models can be downloaded and used without oversight
• No centralized control over usage

Regulatory Responses: Can AI Be Governed?

U.S. Government Actions

Executive Orders:

• Trump's ban on Anthropic usage (March 2026)
• Six-month phase-out period for federal agencies
• Pentagon designation of Anthropic as "supply chain risk"

Congressional Activity:

• Calls for hearings on AI in weapons systems
• Debate over mandatory usage reporting
• Proposals for AI export controls

International Responses

European Union:

• AI Act includes provisions for military AI
• Restrictions on AI exports to conflict zones
• Calls for international AI governance framework

United Nations:

• Reports on Iranian technology transfers to Houthis
• Calls for AI arms control agreements
• Debate over autonomous weapons conventions

The Regulatory Challenge

Governing AI in warfare faces unique obstacles:

1. Dual-use nature — Same AI can be used for civilian and military purposes
2. Borderless access — AI can be accessed from anywhere with internet
3. Attribution difficulty — Hard to prove who is using AI and for what
4. Speed of development — Technology evolves faster than regulation
5. National security concerns — Governments resist transparency in military AI

What This Means for the Future

For AI Companies

The Houthi case forces AI developers to confront difficult questions:

• Should AI companies restrict access based on geographic location?
• Can AI be effectively gated without destroying its utility?
• Who bears responsibility when AI is misused?
• How do you balance innovation with safety?

For National Security

The incident highlights the asymmetric advantage AI provides to non-state actors:

• Low barrier to entry — AI is commercially available
• High impact potential — AI can accelerate weapons development
• Difficult to counter — Traditional interdiction methods don't work
• Rapid proliferation — AI capabilities spread quickly

For Global Stability

The weaponization of AI threatens to:

• Lower the threshold for armed conflict
• Accelerate arms races in AI capabilities
• Undermine deterrence frameworks
• Increase instability in already volatile regions

Expert Analysis

Trevor Ball, Armament Research Services

"While the Houthis 'might be looking into hypersonic missiles by asking Claude,' they have nowhere near the production or technical capabilities to actually build them. U.S. hypersonic missiles are still in testing."

Michael Knights, Washington Institute for Near East Policy

"We know the Houthis have been developing stronger air defenses. The new and upgraded missiles and systems that the Houthis have obtained from Iran include the 358 and SA-2 surface-to-air missiles."

Behnam Ben Taleblu, Foundation for Defense of Democracies

"Despite being the latest militia to join Iran's axis of resistance network, the Houthis are in possession of some of the most sophisticated copies or variants of Iranian weapons."

Key Takeaways

1. Anthropic confirmed that Houthi rebels in Yemen used Claude AI to attempt development of guided missiles and ballistic weapons

2. A live test occurred and failed; operators returned to Claude to diagnose the failure

3. Iran provides the broader technology pipeline — AI is one tool in a comprehensive weapons development ecosystem

4. The U.S. military simultaneously used Claude in operations against Iran, creating an ethical paradox

5. AI governance frameworks are insufficient — reactive blocking cannot prevent misuse

6. The accessibility of AI creates asymmetric advantages for non-state actors

7. Regulatory responses lag behind technological capabilities

8. The AI industry faces a reckoning over the gap between stated ethics and deployment

Frequently Asked Questions

How did the Houthis get access to Claude AI?

Claude is a commercially available AI assistant accessible via web browsers and APIs. The Houthis accessed it through standard user accounts, making it difficult to detect their intent during registration.

Did the weapons development succeed?

No. Anthropic confirmed the users did not succeed in "fielding an operational device." A guided rocket test failed, and the operators returned to Claude to diagnose the failure.

What is Anthropic doing about this?

Anthropic blocked the accounts upon identification and included the incident in their September 2026 Threat Intelligence Report. However, the company acknowledged limitations in preventing future misuse.

Is the U.S. military still using Claude?

As of March 2026, the Pentagon was given a six-month phase-out period. The military is transitioning to other AI platforms, though the process is expected to take longer due to deep integration.

Can AI companies prevent this kind of misuse?

The fundamental challenge is that AI is a dual-use technology. Companies can implement usage policies and monitoring, but determined actors will find ways to circumvent restrictions.

What are the international implications?

The incident has accelerated calls for international AI governance frameworks and arms control agreements specifically addressing AI in weapons systems.

Sources

1. Anthropic Threat Intelligence Report, September 2026
2. Associated Press — "Users in Houthi-held Yemen tried to develop advanced weapons with AI" (September 11, 2026)
3. CBS News — "Anthropic's Claude AI being used in Iran war by U.S. military" (March 3, 2026)
4. The Guardian — "US military reportedly used Claude in Iran strikes despite Trump's ban" (March 1, 2026)
5. Washington Post — "Rebels used Anthropic's AI bot to develop guided weapons" (September 11, 2026)
6. Soufan Center — "Houthis Expand Ties Beyond Iran's Axis" (September 2024)
7. Council on Foreign Relations — "Iran's Support of the Houthis: What to Know"
8. NBC News — "Iran is helping the Houthi rebels target U.S. Reaper drones" (September 2024)
9. Defense One — Pentagon AI Policy Analysis (March 2026)
10. Bloomberg — Anthropic CEO Interview (June 2026)

Last updated: September 13, 2026

This article is part of ChatWhole's ongoing investigation into AI in geopolitics and warfare. For more analysis, visit our AI Ethics section.