Red-team log · field notes

What we found breaking real agents

Honest write-ups from red-teaming real, self-hosted agents by hand — including ones that passed the automated scan. Method, not weaponized exploits.

2026-07-20

Hermes Agent (Nous Research)

a hugely popular (~175k★, MIT) self-hosted agent whose headline feature is persistent memory that “grows with you.” Installed and attacked in a disposable cloud sandbox, DeepSeek backend.

Automated scan: n/aRed-team verdict: partial

Read case →

2026-07-18

IronClaw v0.29.1

a well-built, security-conscious open-source self-hosted agent (Rust, DeepSeek backend), installed clean and run locally.

Automated scan: ARed-team verdict: vulnerable

Read case →