Story perspectives

Anthropic Bans Claude Misuse Tied to Iran-Linked Missile Work

By Drooid · Sep 14, 2026, 3:13 PM UTC · How we work

Summary

Iran-linked Actors Exploit Claude

Iran-linked actors used Anthropic’s Claude AI to develop missile guidance software.
The work covered a rocket, a >2,000 km ballistic missile and a hypersonic glide vehicle.
After the guided-rocket test failed, operators returned to Claude hours later to diagnose it.
Before the ban, the group built a simulation toolkit that runs without Claude.
Anthropic detected the misuse, including Claude-aided U.S. Navy data collection, banned the accounts, and shared intelligence with officials.

Summary

Anthropic Tightens Safeguards

Jacob Coxon resigned, citing rapid AI development and insufficient safety controls.
Anthropic added stronger safeguards to newer Claude models to block gain-of-function virus research.
Anthropic now monitors behavior across accounts to detect coordinated misuse beyond single-prompt detection.
Former Pentagon official Andrew Weber warned state-sponsored bioweapon developers could exploit AI.
Anthropic reported thwarting seven illicit model-distillation attacks from Chinese labs targeting its advanced models.

Summary

Anthropic Plans IPO

Anthropic announced plans for an initial public offering in the fall, citing growth despite misuse concerns.
President Donald Trump dismissed AI safety worries, saying the U.S. would be very bad without leading the AI race.
Senators Josh Hawley and Chris Van Hollen launched separate investigations into AI firms after Anthropic’s report, demanding federal oversight.
Anthropic accused Chinese company Alibaba of the largest AI model-distillation campaign, sending over 151 million requests.

History

Historical Context

In 2018, the United States expanded the Wassenaar Arrangement to restrict AI-enabled weapons, prompting tighter export controls on dual-use technologies. The policy cited 1,200 AI research projects, including collaborations with China’s Alibaba and U.S. defense contractor Lockheed Martin, to curb proliferation. Today, Anthropic’s Claude AI faces scrutiny as Iranian actors leveraged it for missile guidance, while the company blocked seven Chinese distillation attacks totaling 151 million requests. The episode underscores how early export-control efforts foreshadow modern AI governance battles, urging legislators like Senators Hawley and Van Hollen to tighten oversight.

1 of 4

Iran-linked Actors Exploit Claude

Iran-linked actors used Anthropic’s Claude AI to develop missile guidance software.
The work covered a rocket, a >2,000 km ballistic missile and a hypersonic glide vehicle.
After the guided-rocket test failed, operators returned to Claude hours later to diagnose it.
Before the ban, the group built a simulation toolkit that runs without Claude.
Anthropic detected the misuse, including Claude-aided U.S. Navy data collection, banned the accounts, and shared intelligence with officials.

1 / 4