A collection of research and engineering projects for securing LLM-based Web Agents against direct prompt injection, unauthorized actions, privacy violations, and excessive agency. Includes runtime defenses, action verification, security benchmarks, memory protection, and least-privilege controls.
access-control prompt-injection llm-security ai-agents-and-tools llm-security-prompt-injection-ai-safety llm-security-tools web-agent-security
-
Updated
Nov 5, 2026 - Python