SILICON VALLEY, CALIFORNIA (SEPTEMBER 11, 2026) — I still hear conversations about the risks of AI that include the possibility that bad guys could one day use AI chatbots to make weapons of mass destruction.
Sorry to be a Debbie Downer here, but this “eventuality” is already happening, including attempts to develop biological weapons.
This is especially relevant today on September 11, on the 25th anniversary of the terror attacks. The suicide-terrorists on that disastrous day were unwittingly helped at every turn by well-meaning Americans. American flight schools trained the pilots with no background checks or immigration coordination. States issued driver’s licenses and IDs to 14 of the 19 hijackers. U.S. banks processed all their wire transfers without suspicion. And an airline ticket agent even held the flight to ensure two late-arriving hijackers made it onboard.
We’re doing it again, this time with American AI tools.
Anthropic published a report yesterday called “Detecting and Countering Misuse of AI: September 2026.”
According to the report, Yemeni “threat actors” (Houthis, obviously, an Iran-backed terrorist organization aiming to destroy both Israel and the United States), are building weapons using Claude Code.
These weapons include a multi-stage ballistic missile with a range beyond 2,000 kilometers, and a “family” of hypersonic missiles.
The Houthis are using Claude Code to “vibe code” guidance, navigation, and control software for the missiles as well as tune control settings and perform flight simulations.
THE NEW THING: The Gastronomad Pubcast: “Savoring Sicily at the Foot of Mount Etna”
READ: The Attachment Economy, Computerworld, Gastronomad newsletter, Gastronomad book
LISTEN: Superintelligent, TWiT, Gastronomad podcast
JOIN: The Gastronomad Experience
FOLLOW: Gastronomad on Surf Social, Machine Society on Surf Social, Bluesky, Reddit, Notes, Mastodon, Threads, X, Instagram, Facebook, Linkedin!
LOOK AT: Mike Elgan Photography
The group used reinforcement learning to tune flight-control settings, accelerating development of guidance algorithms.
The report says that the Houthi cell test-fired their guided rocket, and the test appears to have failed, after which the actors returned to Claude to diagnose the problem. They’ll keep trying until they succeed.
Anthropic reported on six specific cases in which rogue actors made serious attempts to use AI to develop dangerous conventional weapons: three in China, two in Russia, and one in Yemen.
One group in Russia that is not the Russian government used Claude Code to build an autonomous kamikaze drone swarm.
The drone swarm could select people to kill, and kill them without a “human in the loop.” The group even trained a vision classifier on scraped Ukrainian combat footage.
The company also reported on a China-based actor that used Claude to draft a Chinese-language specification for an anti-torpedo fire-control system.
We also learned that an Iran-linked threat actor used Claude to compile “targeting handbooks” to track US naval positions based on open-source information. The system hoovered up data from publicly accessible ship and aircraft transponder identifiers, commercial satellite-imagery query scripts, and an inventory of public websites that exposed US naval movements. They even harvested information from the captions on public US military photographs.
They also used Claude to compile vulnerability research on shipboard systems.
And in another part of the report, Anthropic revealed attempts to use Claude for potential bioweapons of various kinds, some of them thwarted by Anthropic’s controls, and others not so much.
