Praetorian Releases Augustus: Open-Source Tool for Testing LLM Prompt Injection Vulnerabilities
Key Takeaways
- ▸Augustus is an open-source LLM vulnerability scanner that tests models against 210+ adversarial attacks including prompt injection, jailbreaks, and data extraction
- ▸The tool supports 28 LLM providers and is distributed as a single Go binary for improved performance over Python alternatives
- ▸Recent research shows 86% of production LLM applications are vulnerable to prompt injection, with some attacks achieving 98-100% bypass rates against major models
Summary
Cybersecurity firm Praetorian has released Augustus, an open-source vulnerability scanner designed to test large language models against prompt injection and other adversarial attacks. The tool comes one month after the company launched Julius, an LLM fingerprinting tool, and represents the next step in their AI security toolkit. Augustus addresses what OWASP identifies as the #1 security risk in LLM applications: prompt injection vulnerabilities.
The tool tests LLMs against over 210 adversarial attacks, including prompt injection, jailbreaks, encoding exploits, and data extraction attempts. Built as a single Go binary, Augustus supports 28 LLM providers out of the box and generates actionable vulnerability reports. It was inspired by NVIDIA's garak but reimplemented in Go for improved performance, offering faster execution and lower memory usage than Python-based alternatives.
Praetorian's release highlights a critical gap in the AI security landscape: organizations are deploying LLMs into production environments without adequate adversarial testing. Recent research underscores the urgency, with studies showing 86% of production LLM-integrated applications are vulnerable to prompt injection. The FlipAttack technique achieved 98% bypass rates against GPT-4o, while DeepSeek R1 showed a 100% bypass rate in separate testing. Augustus aims to bridge the gap between rapid LLM deployment and security validation, providing developers and security teams with a tool to identify vulnerabilities before attackers can exploit them.
- Augustus follows Praetorian's release of Julius, an LLM fingerprinting tool, completing a two-step process of identification and security testing
