Hard to say. The current methods are still quite primitive and so relatively easy to be detected.
There is a type of neural networks called Generative Adversarial Networks (GANs) introduced in 2014, initially for image generation, but in the last few years they have been used for many other things (still mostly image manipulation). In layman's terms, those networks generate fake images that look quite real and have a very similar distribution to real images. Nowadays, you can generate beautiful high resolution images which are hard to be discriminated from real ones. They still don't work as good in manipulation as in just generating images, and still don't work as good in videos, but we are getting there. The problem is that when GANs become reliable to do DeepFake videos, there is feck all you can do to discriminate those videos from real ones (again in layman's terms, GANs are actually 2 networks competing against each other, where one tries to generate stuff that looks real, while the other tries to discriminate between fake and real. If the generator wins, then there is no way on knowing that those images are fake, cause the generator is so good at fooling the discriminator. Again, this is simplified, but hopefully you got the idea).
Anyway, we are probably a few years from it happening, and many people are working on this, so someone might find new ways of being able to detect fake from real. It is both exciting and scary.