Prompt Guard expands detection for diverse prompt injection attacks, adding 6 new categories and over 500 patterns.
No data available
Released v2.7.0 with detection capabilities for 6 new attack categories including auto-approve exploitation, MCP tool abuse, Unicode tag injection, pre-filled URL exfiltration, browser agent injection, and log viewer context exploitation, bringing total blocked attack patterns to over 500.
Added multilingual prompt injection detection support for 10 languages including Russian, Spanish, German, French, Portuguese, and Vietnamese with over 60 new patterns in v2.6.2.
Implemented detection for system prompt mimicry, GODMODE/DAN keywords, and leetspeak evasion techniques in v2.5.1.
Added detection for BRC-20 JSON injection, sovereignty manipulation attacks, scope creep exploitation, credential path hiding, and DM-based social engineering through versions v2.5.2 to v2.6.1.