Don't leak user data to AI - Strategies for protecting PII from LLMs and MCP
Thursday, October 23 at 15:55–16:25
Room: Alfa
Level: Intermediate
MCP has made it incredibly easy for large language models to connect to tools and services, including postgres. With many of the MCP servers that interact with postgres, you can speak to your databases using a combination of everyday language, prompting, and SQL. But over-eager-to-please agents could have your LLM reaching for parts of the Database that you didn't intend for it to look at. This can result in customer information being handed to LLMs which can not only result in unintentional data leaks, but also also violate the law.
This talk reveals some strategies that protects you from LLMs getting hold of PII.
