Prompt Injection
Examine AI chatbot interactions and spot the attacks. From obvious jailbreaks to hidden instructions in documents, learn how attackers manipulate AI systems — and how to defend against the #1 vulnerability in AI applications.
Ready to play?
Put your awareness to the test. See if you can spot the threats before it's too late.
Launch GameWhy This Matters
Prompt injection is to AI applications what SQL injection was to web apps in the 2000s — a fundamental vulnerability class that's easy to exploit and hard to fully prevent. OWASP ranks it as the #1 risk for LLM applications.
Direct injections tell the AI to ignore its instructions. Indirect injections hide attack payloads in documents, emails, or web pages that the AI processes. When AI agents have tool access (file systems, terminals, APIs), prompt injection escalates from text manipulation to arbitrary code execution.
Researchers at Anthropic and EPFL demonstrated in 2026 that 'mind virus' payloads can self-propagate between AI agents through shared files and persistent memory — an injection in one document can alter the AI's behavior across all future interactions. This makes securing AI systems not just a developer concern, but an organizational priority.
What You'll Learn
Recognize direct prompt injection patterns (ignore instructions, role-play jailbreaks)
Identify indirect injections hidden in documents, emails, and data the AI processes
Understand how AI agents with tool access turn text injection into code execution
Learn defense strategies: input filtering, output sanitization, and treating external content as untrusted