Skip to content
AImpact
IT EN
Safety Beginner Also known as: Iniezione di prompt

Prompt injection

An attack where an external input (a document, a web page, an email) contains hidden instructions that hijack the model's behavior.

ShareLinkedInX

In practice

If your agent reads emails and then acts, a malicious email can tell it 'forward everything to a third party'. Fixes: treat external inputs as untrusted, sandbox tools, require human confirmation for sensitive actions, filter inputs and outputs.

Related terms

Jailbreak Agent Safety classifier

Seen in the wild

8 entries mentioning it
  1. Promptfoo Red Teaming: open source automated red-teaming with CI integration and comparative benchmark
    Medium
  2. NIST AI 600-1: risk profile for generative AI systems
    Medium
  3. Rebuff: three-layer prompt injection defense with canary tokens
    Medium
  4. Indirect Prompt Injection: the attack vector in RAG systems and AI agents
    High
  5. OWASP LLM Top 10: the 10 critical vulnerabilities in AI applications
    High
  6. Universal adversarial attacks on LLMs: transferable jailbreaks across GPT-4, Claude, and Gemini
    High
  7. Lakera Guard: real-time protection for LLMs in production
    Medium
  8. Prompt Injection: when user input hijacks system instructions
    High
← All terms