51 papers out of 4841 accepted papers is 1%. If these were the only papers, then I would say the review system actually doing pretty well. Considering how noisy the review processes are, I don't see this as a big problem on its own.
We can't stop people (or agents) from submitting low quality papers in the current system. We can improve the review quality, but at a human cost.
Foundation models can be seen as approximate amortized posterior inference machines where the posterior is conditioning on the pre-training data. However, the uncertainty is usually ignored, and there may be ways to improve the state of the art if we were better Bayesians.