“The concern is no longer that someone asks a chatbot to write malicious code.”

Someone used Claude to debug missile guidance software. Let that sink in for a second.

What the report found

Anthropic published its September 2026 threat intelligence report this week, and it is a lot to take in. Between December 2025 and August 2026, its team identified and disrupted misuse of Claude across seven harm areas: cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and distillation.

Cases that read like a geopolitical thriller

A China-based actor built an electronic warfare and air-defence suppression suite, with a default scenario that shifted mid-project to 12 targets in Taiwan. A group in Yemen used Claude to develop guidance software for a rocket, and when the test firing failed, they came back to Claude to debug it. A freelance team in Russia used Claude Code to build an autonomous kamikaze drone swarm.

The shift that matters most

The thing that hits me hardest is not the ambition of the attacks. It's what Anthropic said about the shift in how AI is being misused. The concern is no longer that someone asks a chatbot to write malicious code. The actors here used AI to take over multiple steps of an operation, end to end.

What responsible disclosure looks like

Anthropic did the right thing by publishing this. It disrupted the activity, shared intelligence with authorities and industry partners, and built new classifiers specifically for weapons development traffic. That is what responsible disclosure looks like.

To be fair, none of this is unique to Claude. State-sponsored groups have been caught misusing Google's Gemini as well, and the broader research community has documented this dual-use problem clearly. AI has collapsed what Anthropic calls “the labour and tooling gap” that used to separate well-resourced state actors from individual operators. That is the uncomfortable truth underneath all of this.

I've spent well over 20 years watching technology get weaponised in one way or another. This is different in scale and speed, and I don't think we're close to solving it.

If you're connecting AI agents to your own systems and want a second set of eyes on where the oversight gaps might be, that's exactly the kind of conversation worth having. Find me through the contact section of this site.