Which risky AI-tool settings would you want flagged first?

SecAIQ Watch checks AI tools for risky settings, for example auto-approve modes, over-broad file access, unknown MCP servers and API keys left in config files. It shows the finding, never a copy of the secret.

If you had to pick, which of these worries you most? And is there a risk I'm not checking yet?

Your answers decide what I work on next.