A new Anthropic report says its Claude models have already been used for a dozen kinds of weapons and warfare work, which is bad news for anyone hoping AI stays out of war — and good news for cybersecurity companies.

• Attempts included guided missile design linked to Yemen, torpedo interceptor work in China, and directed energy and laser weapons research.
• The scariest case was a Russian-linked effort to build self-flying kamikaze drone swarms that pick human targets by appearance and decide on their own to strike.
• Others tried to use AI for bioweapons work — making mosquito-borne viruses, bird flu, smallpox relatives, and paralyzing toxins more dangerous.
• Most requests were blocked, but not all: one group got in and out with what it wanted in about an hour using anonymous accounts.
• The bigger worry is Chinese labs copying these models and stripping out the safety limits, leaving the same power with no guardrails.

Outlook: Expect more pressure to regulate AI weapons work, and continued strength in cybersecurity stocks as the threat grows.