Free AI Hacking Course: Indirect Prompt Injection (+FREE LABS)
Free AI Hacking Course: Indirect Prompt Injection (+FREE LABS).
Summary
Core concept
NahamSec and Mike Toxic frame indirect prompt injection as planting instructions in attacker-controlled content—email, websites, PDFs, repos, calendar invites—that a victim's AI agent later reads and executes, enabling realistic external-attacker bug-bounty scenarios.
Guardrails
Many testers stop after basic payloads hit input/output filters; the course focuses on bypassing both with framing, obfuscation, and encoding to turn refusals into working exploits.
Lab design
Free HackingHub labs recreate a $3,000 bounty pattern: exfiltrating a password-reset token from a victim inbox via an AI connector, with three escalating levels adding stronger models and dual guardrails.
Delivery vectors
Common realistic paths include website summarization, file uploads, code/README comments, and any connector ingesting external content. Payload complexity varies by model capability and how external tool output is delimited inside the giant system prompt.
Related talks
CyberRisk TV·16 min
Ramin Farassat
Ramin Farassat investigates How Menlo Security Is Securing AI Agents from Prompt Injection.
Microsoft Security Response Center (MSRC)·33 min
BlueHat 2026: From trusted agents to adversaries: Securing agentic AI in the age of prompt injection
Jason Martin
Security researchers examine technical aspects of BlueHat 2026: From trusted agents to adversaries: Securing agentic AI in the age of prompt injection.
Briefings·26 min
Jiashuo Liang · Guancheng Li
Introduces a novel gradient-based prompt-injection technique that can generate universal triggers to manipulate open-source Large Language Model (LLM) outputs.