AI created two risks older tools miss: people pasting secrets, PII and code into AI apps, and attackers injecting your models to extract context. Palivane inspects both directions everywhere AI touches your org: a browser extension, an OpenAI/Anthropic/Gemini compatible gateway, connectors for Slack, M365 and Workspace, plus CLI and CI hooks. Detection runs cheapest first. Regex rules, then two on-box ML classifiers, then an optional LLM judge that runs under your own cloud contract.