Evaluating Safety Embedding Prefiltering for Analyzing Millions of LLM Agent Social Network Messages for Security and Safety Harms

Summary

The study examines whether lightweight embedding-based prefilters can cut the cost of screening agentic social network content for security and safety harms. Researchers annotated 10,000 Moltbook posts and comments with a frontier LLM judge, finding 9.2% unsafe at severity 3 or above. At 0.80 recall, prefiltering reduced the projected cost of scanning 787,226 messages by 49.9% to 65.6%, depending on the encoder or trained classifier, though jailbreak content remained the hardest category to retrieve.

Related items

• CriticalHermes Agent - PKCE Session Takeover via Redirect-URI Parser ConfusionSimilar attack · Tenable Research Advisories
• LowSocial Engineering AI Agents: The New BEC for 2026Similar attack · Dark Reading
• LowLost in the comments: Social context as a single‐pass jailbreak and defense on agentic platformsSimilar attack · OpenAlex (peer-reviewed AI security)
• MediumGHSA-hmq2-7hp6-7crh: Banks: User-controlled prompt input can be parsed as privileged chat messagesSimilar attack · GitHub Advisory Database
• HighGHSA-6wjp-v33h-5cvq: PraisonAI: AgentOS defaults to network-exposed no-auth mode, allowing unauthenticated agent invocation and instruction disclosureSimilar attack · GitHub Advisory Database