The Data Problem: Why LLM Security Is So Complex

The Data Problem: Why LLM Security Is So Complex

Large language models are trained on terabytes of data, but what happens when that data is flawed? In this video, A10 Networks' security experts, Jamison Utter, Madhav Aggarwal, and Diptanshu Purwar, discuss a critical, often-overlooked aspect of AI security: the training data itself.

They explain that LLMs are inseparable from the data they're trained on, which means if the data contains biases, toxic content, or other vulnerabilities, those flaws are vulnerable to exploitation by attackers.

These issues, combined with the sheer volume of data, create a vast and complex attack surface. The experts highlight that a single security control is insufficient to mitigate this "barrage of attacks" during inference time, as the attacks can come from multiple, unexpected vectors, posing a new challenge for cybersecurity teams.

Discover why the data an LLM is trained on is a primary source of security risk and how organizations can address this new frontier in AI security.

Learn how to secure AI and LLMs in your organization: https://bit.ly/4kOHmYd.