That's an important question.
I think it's important to start by just noting what we have seen with regard to AI-generated content from foreign interference operations so far, and here I would note that the use of AI-generated content is not new. Since 2019, we have identified CIB operations on our platforms that have used profile pictures generated by a technique called generative adversarial networks, also known as GAN profiles. This use of AI-generated content doesn't allow these networks to evade our behaviour-based detections. In fact, over two-thirds of the CIB operations that we removed in the last year featured this type of AI-generated content.
We have seen newer operations using the latest generative AI techniques, and there are challenges that we anticipate there, including related to the scale of the content that can be created, but we fundamentally believe, from what we have seen, that a behaviour-based approach is still well suited for identifying covert influence operations early in their life cycle.