Safeguarding against Data Exfiltration Risks in the Era of AI-driven Conversational Analytics

AI-driven conversational analytics has huge potential to finally deliver on the self-serve dream that has been promised since the early days of BI. It has the potential to significantly reduce time to insights, TCO and dashboard development backlogs.

But submitting raw data and contextual information to LLMs must be done in a controlled and well-governed fashion. Not only must user prompts only be augmented with data that is approved for submission to a model, but the model must also ensure it only processes and returns data that the user has legitimate access to.

In this session, we will look at the different layers of governance that are needed and how they must span physical deployments, LLM access controls and strict data compartmentalisation capabilities in order to prevent unauthorised data exfiltration.

Key takeaways:

• Considerations for data security within AI-driven analytics
• How to implement strict cross-platform data compartmentalisation policies
• How data compartmentalisation can lead to non-deterministic result sets and how to manage this