AGENT-EGRESS-BENCH

Summary from the TLDR InfoSec newsletter

agent-egress-bench is a standardized test corpus for evaluating AI agent egress tools, covering secret exfiltration, prompt injection, SSRF, hostname exfiltration, MCP tool poisoning, chain detection, MCP drift, A2A protocol scanning, WebSocket DLP, encoding evasion, shell obfuscation, and cryptocurrency/financial data protection.