How can you mitigate prompt gaps and leakage?

• A. Do not attempt to address context; rely on model memory.
• B. Increase model size to compensate the lack of context.
• C. Provide explicit context, use checklists, constraint handoffs, and content filtering, with careful context scoping.
• D. Remove system prompts to simplify prompts.

Answer: (C)

Mitigating prompt gaps and leakage comes from giving the model clear, bounded context and guardrails rather than hoping it will recall everything or behave safely on its own. Providing explicit context means including the goals, constraints, and any relevant data in the prompt so the model doesn’t have to guess or fill in missing pieces, which reduces both errors and stray, unintended outputs. Using checklists helps ensure you cover all the required steps, decisions, and safety checks in a repeatable way, preventing important details from being skipped. Constraint handoffs clarify responsibilities and boundaries—what the model should do, what it shouldn’t do, and when to escalate if uncertainty arises—so outputs stay within acceptable limits. Content filtering acts as a safeguard to block or flag problematic content before it’s produced, further reducing leakage risk. Careful context scoping means including only the information that’s truly needed and relevant, avoiding overload that can confuse the model or reveal sensitive data.

Relying on the model’s memory rather than explicit context is risky because memory is not reliable across sessions and prompts. Simply increasing model size doesn’t fix gaps or leakage, since understanding and safe behavior still depend on how the prompt is framed. Removing system prompts removes guiding instructions that help constrain behavior, increasing the chance of inconsistent or unsafe outputs.