ADL study finds popular open-source LLMs can easily be manipulated by malicious actors to produce antisemitic, extremist, and dangerous content amid weak safety guardrails.

The first analysis of hate and harassment on Minecraft server data.