A gaming company is deploying a public chat companion built on a Foundry-hosted model. Before launch, the team must reduce the risk that the companion could be tricked into producing hateful, violent, or self-harm-related content. Which Foundry capability should they configure as part of the deployment?
Select an answer to reveal the explanation.
Short Explanation
Picture Content Safety as a bouncer standing at the door of every response before it reaches a player, checking it against categories like hate, violence, and self-harm and stopping anything that crosses the line. That is the piece built specifically for this worry. Making responses longer does nothing about what is in them, so a bigger token limit is not a safety control at all. Forcing every conversation down one identical scripted path would solve the harm problem only by making the companion pointless to talk to, which is not what harm mitigation actually looks like in practice. And picking a model because it answers quickly, based on catalog benchmark numbers, tells you nothing about whether its answers are safe; speed and safety are measured by entirely different systems within Foundry.
Full Explanation
The correct answer is A. Content Safety filters in Microsoft Foundry are designed to score both prompts and model outputs against harm categories such as hate, violence, sexual content, and self-harm, and can block or flag content that crosses configured severity thresholds, which directly addresses the company's concern about the companion producing harmful material. Option B is incorrect because a higher token limit only changes how long a response can be; it has no effect on whether that response contains harmful content. Option C is incorrect because routing every conversation through one fixed reply would make the companion useless for normal interaction and is not how harm mitigation is implemented; safety is applied as a check on generated content, not by eliminating variation entirely. Option D is incorrect because benchmark scores in the model catalog describe quality, cost, and latency trade-offs between candidate models; they do not measure or reduce the risk of harmful output from a deployed model. Configuring Content Safety filtering as part of the deployment gives the team an ongoing, adjustable safeguard rather than a one-time model choice.