Forget relying on fancy AI "guardrails" and model providers to keep things secure as AI automates more! A privacy expert reveals how easy it is to bypass current protections, thanks to models memorizing sensitive data and architectural quirks.
Engineers need to ditch one-off red-teaming and the idea that the next AI version will magically fix everything. Instead, they should build a culture of security, embrace iterative threat modeling, and consider local LLMs or diversifying providers for better safety.
Watch on YouTube
Top comments (0)