Case Studies

LFI, path traversal, SSRF chains — reading files and secrets the app never meant to expose.

• └──Trusted Chat to ZIP Upload with Symbolic Links8 min read
• └──Trusted Chat to Pickle Deserialization via bashExecutor10 min read

Deserialization, template injection, unsafe pipelines — getting a shell where there shouldn't be one.

• └──Trusted CSV Chat to Remote Pickle Deserialization9 min read
• └──Malicious Python Artifact Executed by an AI Sandbox10 min read

Client-controlled context and message roles that blur the boundary between user content, assistant output and system instructions.

• └──Prompt Injection via Chat Selection and Role Confusion7 min read

Classic and second-order injection — the bug class that never dies, even in modern stacks.

• └──RAG Debug Response to Cross-Tenant SQL Injection12 min read

Testing whether hidden instructions, policies and system context can be extracted from AI-powered applications.

• └──System Prompt Exfiltration Through Agentic Workflow Creation4 min read