A new Anthropic report lays out a dozen ways AI is already being turned into a killing tool, which is bad news for anyone hoping safety rules are holding.

• Suicide drones are being wired up with AI to spot people and tag them as friend or enemy by appearance.
• China is working AI into torpedo systems, and groups in Yemen are using it to bolt phone-grade guidance onto long-range missiles.
• AI is also being used to study viruses for gain-of-function work, including bird flu strains that kill a large share of the people who catch them.
• Safety guardrails are leaky: model outputs get copied into cheaper student models, and those copies can have the protections stripped out.

Outlook: Expect more pressure on AI companies and governments to police military and biological uses, but the tools are already out in the wild.