A large-scale cross-framework agentic security evaluation by an unnamed research team found that orchestration framework choice explains only 0.06% of outcome variance, while attack family accounts fo…
AgentSafe Labs retracted a previous finding about Claude Haiku returning UNCERTAIN results in red-teaming tests, revealing the cause was a bug in its own safelabs-eval detector logic rather than ambig…
OWASP's Agentic Security Initiative Top 10 (ASI01-ASI10) provides a threat taxonomy for AI agents that use tools, memory, and multi-agent communication, distinct from the LLM Top 10. Testing 30 advers…
AI security defenses built for chatbots fail against prompt injection in agentic systems because agents have multiple input channels—documents, tool outputs, memory—that are not monitored or filtered.…
AgentSafeLabs tested Claude Haiku against prompt injection attacks, with two of three ASI01 tests passing and one returning UNCERTAIN. The UNCERTAIN result highlights challenges in defending agentic s…