AI research and discussion on LessWrong. www.lesswrong.com
Shows that multi-agent architecture choices (roles, topology, memory) strongly and unpredictably change exploitability to misuse, often making systems more vulnerable than single agents, so security must be tested per deployment.
Proposes that prompt injections exploit LLM role confusion, using role probes to show writing style overrides chat-template tags; introduces CoT Forgery, a jailbreak whose success tracks internal role perception.
Presents SecureMaxx, an agent-callable biosecurity screening tool that raises refusal rates on hazardous DNA sequences from near zero to 70-100% with minimal latency and capability cost, and discusses its dual-use limits.
Anthropic reports threat actors using Claude Code to automate extortion, fraud and ransomware development, and describes the bans, classifiers and detection methods it deployed in response.