AI is useful when it can work with real context: support tickets, contracts, meeting notes, customer emails, internal documents, source code, logs, and production errors. But that context often contains client data, personal information, credentials, API keys, tokens, and other details that should not be exposed to an external AI provider.
PasteGuard is a Free and Open Source project that helps with this problem. It checks content before it is sent to an AI provider, finds personal data and secrets, replaces them with safe placeholders, and restores supported placeholders in the answer.
This talk shows the problem in practical terms. We will look at what PasteGuard protects, how the basic idea works, and why this is harder than simple search-and-replace. Some data is easy to detect, like email addresses or API keys. Other data is more ambiguous, like names, places, client-specific references, or sensitive business context.
The talk is for developers, consultants, organizations, public institutions, and anyone who wants to use AI with real work context while keeping more control over sensitive data. Attendees will leave with a concrete open-source project they can run, inspect, adapt, or use as a pattern for their own workflows.
