19Users say our assistant refuses harmless requests; legal fears it will answer harmful ones. How do you evaluate this?▼mediumNewAnthropicOpenAIMeta◆ premiumTwo teams pulling in opposite directions, each with a real point. The strong answer gives them one shared measurement so the argument becomes a decision.Practise this question →Read answer →
31Our PMs' prototypes install whatever packages the coding agent suggests. Security wants to ban PM prototyping. What do you propose instead?▼mediumNewReplitAnthropic◆ premiumModels sometimes invent package names, and attackers register them. Keep prototyping, but allow-list common libraries, check that new packages exist and are established, and install in a sandbox with no secrets.Practise this question →Read answer →