Practice question 42 of 498 · hard
A previously evaluated AI assistant gains a new external-data connector and write-capable tool. Which two actions should precede wider release? Select two answers.
Select all that apply. Every correct option and no extras are required for credit in the app.
Show the answer and explanation
✓ Correct answers
- Repeat adversarial tests covering the new content and tool paths
- Revalidate tool permissions, data boundaries, monitoring, and approval requirements
Repeat adversarial tests covering the new content and tool paths: New inputs and capabilities change the attack surface and can invalidate old findings. Revalidate tool permissions, data boundaries, monitoring, and approval requirements: Safety depends on enforcement and visibility across the changed application. Reuse the original evaluation without considering the new connector: The original assessment did not cover the changed input path. Assume a trusted model makes every integration safe: Integration permissions and content introduce independent risks. Remove the old restrictive controls to simplify the new tool: Simplification must not silently broaden authorization.
Read and review here, or open the app to practise and save your results on this device.
Unofficial study material. Not affiliated with or endorsed by Microsoft. These are original learning questions, not actual exam questions. Product capabilities and exam objectives can change; verify current Microsoft documentation. Practice scores do not predict or guarantee an exam result.