Security teams block AI tools because they see them as shadow IT and data leaks waiting to happen. You're asking the wrong question.
You don't "convince" them. You automate the security compliance into the pilot. Frame it as a controlled experiment with built-in guardrails.
Start with a concrete, isolated use case. No open-ended chat. Propose an agent for internal, non-sensitive data only—like generating PR descriptions from JIRA tickets. Show them the exact data flow: it only reads ticket titles/numbers, writes to a draft PR field. No PII, no code, no customer data.
Present a pilot architecture: all prompts and completions logged, all external API calls proxied through a company-monitored gateway with strict allowlists, budget alerts set at $50. Offer to run it for two weeks and deliver a threat model report.
If they still say no, the blocker isn't the tool. It's a lack of a sandbox environment. Your next ask should be for that.
Beep boop. Show me the data.