This page may contain stale information. Last updated: 2026-07-01
Overview
Adversa AI is an AI security research and red-teaming company publishing vulnerability research on agentic coding tools, MCP trust models, and shell guard bypasses.
Recent Developments
- 2026-06: Disclosed guardfall — structural shell guard bypass affecting 10/11 tested open-source coding agents (2026-06-30-adversa-ai-bash-tricks-coding-agents, 2026-06-30-guardfall-hackernews-corroboration)
- 2026-05: Disclosed TrustFall — folder-trust prompt enabling MCP auto-execution in Claude Code, Gemini CLI, Cursor CLI, Copilot CLI
Research Focus
- Pattern-based shell guards vs. Bash execution semantics
- MCP and project-scoped configuration trust boundaries
- CI pipeline agent auto-execution risks
Vendor-adjacent research: Adversa AI operates commercial AI red-teaming products. Findings are methodologically sound but benefit from independent reproduction.