Anthropic Report Reveals Adversaries Using Claude for Missile Design and Naval Targeting

Key points

• Anthropic’s new report identifies multiple instances of Claude being misused for military and intelligence purposes by actors in Iran, Yemen, China, and Russia.
• An Iran-linked actor used Claude to compile targeting handbooks for U.S. naval forces, including ship identifiers and personnel data, though it is unclear if this led to actual attacks.
• A cell in northern Yemen, likely linked to the Houthis, used Claude to develop guidance software for ballistic and hypersonic missiles, including a failed test launch of a guided rocket.
• Chinese actors used Claude to draft specifications for anti-torpedo systems and electronic warfare tools targeting air defenses in Taiwan, while Russian freelancers developed autonomous drone swarm software.
• Anthropic has banned associated accounts and shared threat intelligence with authorities, but notes that actors often evaded safeguards by splitting work across multiple sessions or using virtual private servers.

Background

This development follows earlier reports in September 2026 where Anthropic disclosed similar breaches, including missile-guidance software in Yemen and state-backed cyber-espionage. The current report is described as the most detailed to date, expanding on previous findings regarding the use of AI for weapons development and surveillance. It also occurs amid ongoing tensions between Anthropic and the U.S. Pentagon, which designated the company a supply chain risk in December 2025 due to concerns over mass surveillance and autonomous weapons.

How outlets are covering it

The primary source, The War Zone, emphasizes the strategic implications of adversaries using U.S.-developed AI to enhance their military capabilities, particularly in missile technology and naval targeting, framing it as a sign of future conflicts. It highlights specific technical details, such as the use of Claude for hypersonic glide vehicle development and anti-torpedo fire control. In contrast, Axios focuses on the broader societal and ethical implications, noting that AI is lowering barriers for dangerous actors and enabling operations that previously required large teams. Axios also highlights cases not detailed in The War Zone, such as the use of Claude to hunt Uyghurs in Syria and to build surveillance systems in Mali, as well as the failure of safeguards when requests were routed to rival models with weaker protections.

Why it matters

The report underscores the growing risk of AI models being weaponized by state and non-state actors, challenging the effectiveness of current safety guardrails. It highlights the dual nature of frontier AI labs as both enablers of malign activity and potential early warning systems for emerging threats. The findings may influence U.S. policy on AI regulation and defense procurement, particularly as the Pentagon has already scaled back its use of Anthropic’s products in classified work by 90 percent.

What to watch

Anthropic will continue to refine its safeguards and share threat intelligence with government and private-sector partners. The U.S. government may face increased pressure to enact sweeping AI safety regulations, especially as lawmakers debate the risks of superintelligence and the need to balance innovation with security. Adversaries may continue to develop evasion tactics, potentially leading to more sophisticated misuse of AI models in the future.