---
title: '12 Ways AI Is Being Used to Kill People'
source: 'https://youtube.com/watch?v=jKs3DyfRjjw'
video_id: 'jKs3DyfRjjw'
date: 2026-09-11
duration_sec: 176
channel: 'Meet Kevin'
---

# 12 Ways AI Is Being Used to Kill People

> Source: [12 Ways AI Is Being Used to Kill People](https://youtube.com/watch?v=jKs3DyfRjjw)

## Summary

Anthropic's report reveals 12 ways AI is being weaponized, from autonomous drones to bioweapons research. The transcript details specific examples, including Russian and Chinese military applications, and highlights the paradox of AI guardrails being removed through distillation.

### Key Points

- **Autonomous Kamikaze Drones** [00:13] — Kamikaze drones programmed with Claude can autonomously label and target people as friendly or enemy, and issue detonation commands. This was done through a Russian-based operation.
- **Missile and Torpedo Development** [00:41] — China-based operations are developing torpedo and anti-torpedo weapon systems. Houthis or people in Yemen are using AI to design homing guidance for ballistic missiles, potentially hypersonic, to fly 2000 kilometers.
- **Supply Chain Attacks and Directed Energy Weapons** [01:07] — AI is used to plan supply chain interruptions and attack enemy component manufacturing. Directed energy weapons, including vehicle-mounted high microwave weapons, are being developed to counter drones and for attack.
- **Bioweapons and Gain-of-Function Research** [01:35] — Viruses are being evaluated for gain-of-function research, including Chikungunya virus (severe pain/fever for weeks/months), killer bird flu variants (fatal in ~50% of confirmed human cases), and Orthopox (smallpox/mpox variant). Anthropic blocked a user researching Orthopox after just one hour.
- **Cyber Threats and Model Distillation** [02:29] — AI is used for exploitation tests to find computer vulnerabilities, agentic surveillance (swarms of agents collecting credentials), and model distillation, which copies AI outputs to teach students cheaply, potentially removing guardrails.

## Transcript

Claude just revealed there are 12 ways AI is being used to basically kill people and it's all being driven by, well frankly, humans. So, it's not clear, are humans the problem or is AI the problem?
But, look at this. Kamikaze drones are getting programmed with Claude to automatically label and target people as friendly or enemy based on what they look like.
And they can issue the command autonomously to detonate against opposing drone operators or people identified as enemies. This was actually done through a Russian-based operation using CLOD models.
But it's not just that. It's China-based operations coming up with torpedo and anti-torpedo weapon systems. It's Houthis potentially or people in Yemen and a weapons engineering cell using AI to figure out how they could use phone computers with homing guidance attached to ballistic missiles, potentially hypersonic missiles to fly 2000 kilometers.
Or we have right here supply chain interruption for how to actually get the components or attack enemy component manufacturing systems for these weapons using AI to figure out how to build directed energy weapons including vehicle high microwave weapons for countering drones as well as for attack purposes
You've got viruses being evaluated for gain-of-function research. Yeah, kind of like what happened with COVID, except now add AI, including the Chikunga Guna virus,
which apparently can cause severe pain and fever lasting for weeks or months. Killer bird flu variants that are fatal in roughly half of confirmed human cases in this highly pathogenic research that's being conducted.
Orthopox, which is a variant of smallpox and mpox, which that research was conducted in just one hour before Anthropic was able to block the people from collecting the information.
toxins that could debilitate people and paralyze them are being researched. Exploitation tests to basically exploit vulnerabilities in computers.
Agentic surveillance, think swarms of agents being used to collect and surveil you or collect your credentials. And then, of course, while Claude is talking about, hey, we're going to come up with guardrails, the reality is Claude is also getting distilled, which basically means copied almost,
where the outputs were getting taught to students cheaply and efficiently at a low cost, which could then potentially remove the guardrails. This entire report from Anthropic is insane.