Praetorian — Augustus open-source LLM prompt-injection scanner
AI relevance: Augustus is a dedicated red-teaming scanner for LLM applications, directly targeting prompt-injection, jailbreak, and data-exfiltration weaknesses in agent and RAG stacks.
- Praetorian introduced Augustus, an open-source LLM vulnerability scanner built as a single Go binary.
- The tool targets prompt injection, jailbreaks, encoding tricks, and data-extraction probes across LLM apps.
- Augustus ships with 210+ adversarial probes and claims out-of-the-box support for dozens of LLM providers.
- Praetorian positions it as a faster, dependency-light alternative to Python-based scanners like garak.
- The release is framed around a deployment/testing gap: teams ship LLM features faster than they red-team them.
- Augustus is open-source and intended to fit into standard penetration-testing workflows.
Why it matters
- Prompt injection remains the top LLM app risk, but operational testing tooling is still sparse.
- Portable scanners lower the barrier to continuous red-teaming for LLM apps in production.
What to do
- Run automated injection suites as part of CI/CD for LLM features, not just one-off audits.
- Compare scanners (Augustus vs garak) to ensure coverage across jailbreaks, encodings, and data leakage.
- Track findings to specific prompts, tools, and RAG sources so fixes are actionable.