AI and LLMs · Senior · Question 23 of 75
What is the "lethal trifecta" for prompt injection in agents?
- 1A large model, a long prompt and a high temperature together
- 2Three failed logins followed by a password reset request
- 3Private data access, untrusted content and a way to send data out
Show the answer
Correct answer: Private data access, untrusted content and a way to send data out
Explanation
When an agent has all three, an injected instruction can read secrets and exfiltrate them. Removing one leg is the most reliable defence.