RESEARCHERS HIGHLIGHT HOW POISONED LLMS CAN SUGGEST VULNERABLE CODE

Summary from the TLDR InfoSec newsletter

There is new research that reveals AI code assistants can be manipulated to suggest vulnerable code. A technique called CodeBreaker can poison training data, creating undetectable malicious code samples using comments in code.