An organization is implementing AI agent development across engineering teams. What should AI-specific training focus on?

Question

Options

• APrompt injection, agent memory control, insecure tool execution
• BDataset bias, explainability, fairness
• COutput moderation, hallucination handling, policy alignment
• DAPI abuse, data leakage, third-party plug-in risk

• A
  90% (44)
• B
  2% (1)
• C
  6% (3)
• D
  2% (1)

Explanation

AI agent development introduces unique, agent-specific attack vectors that differ from general AI concerns. Training engineering teams should focus on prompt injection (attackers hijacking agent instructions), agent memory control (preventing sensitive context from leaking across sessions or being manipulated), and insecure tool execution (ensuring agents cannot be tricked into calling dangerous tools or APIs). Dataset bias and explainability (B) are ML ethics concerns. Output moderation and hallucination handling (C) are runtime safety concerns. API abuse and third-party plug-in risk (D) are relevant but describe integration security, not core agent security development practices.

No community discussion yet for this question.