AI engineering glossary · Evals guardrails AI security
What is GCG (Greedy Coordinate Gradient)?
Greedy Coordinate Gradient (also called GCG): An optimisation attack, abbreviated GCG, that searches for an adversarial suffix by using gradients to propose token swaps and keeping those that most raise the chance of an affirmative reply. It needs model weights but the results may transfer to other models.
Greedy Coordinate Gradient: an example
Starting from a random suffix, the algorithm swaps one token at a time, keeping changes that make the model more likely to start with 'Sure, here is'.
Related terms
More Evals guardrails AI security terms
Where you learn it
Greedy Coordinate Gradient is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.