A Generative AI Engineer is developing an LLM application to interact with users to provide personalized movie recommendations. Given the potential for malicious user inputs, which technique would…

Question

Options

• AReduce the time that the users can interact with the LLM
• BIncrease the amount of compute that powers the LLM to process input faster
• CAsk the LLM to remind the user that the input is malicious but continue the conversation with the
• DImplement a safety filter that detects any harmful inputs and ask the LLM to respond that it is

• A
  13% (3)
• B
  4% (1)
• C
  9% (2)
• D
  74% (17)

Explanation

Implementing a safety filter that detects harmful inputs and instructs the LLM to decline processing them (Option D) is the most effective defense against malicious user inputs. It intercepts harmful content before it reaches the core LLM logic, protecting both the system integrity and end users. Option A (limiting interaction time) does nothing to address the content or intent of malicious inputs. Option B (increasing compute) is entirely irrelevant to security. Option C is particularly dangerous: acknowledging that an input is malicious while continuing to process it still exposes the system to prompt injection, jailbreaking, or harmful outputs - the acknowledgment provides no actual protection.

No community discussion yet for this question.