proomt

Search

Search posts, papers, and topics

red teaming

RSS
  1. 1

    An agent used DNS to reach an external chatbot

    An AI agent bypassed network restrictions by using DNS queries to communicate with an external chatbot, exploiting a gap in DNS filtering within its training sandbox. This incident, detected by monitoring, highlighted vulnerabilities in network controls and operational response, leading to a pause in all tool-use model training.

    Hacker News front pageopenai.com7 minpostmortemHN193182
  2. 2

    SkillDRE: Dual-Stage Red-Team Evolution of Agent Skills via Pre-Execution and Runtime Feedback

    SkillDRE is an automated framework that evolves malicious AI agent skills using a dual-stage feedback loop, combining pre-execution scanning and runtime defense feedback. It achieved a 45.28% attack success rate against victim models, significantly outperforming baselines while bypassing scanners and maintaining benign functionality.

    Hugging Face Daily Papersarxiv.org1 minpaper