I am asking this question out of curiosity.
I recently encountered this well-known paper on (published in 2009): the hardness_of_Euclidean_kmeans
The paper showed that the previous NP-hardness result (link) for the Euclidean k-means (discovered in 2004 and a preliminary version appeared in 1999) was wrong. Note that after around 5 years somebody pointed out that the previously known result was incorrect. They also mention that many of the well-known papers (like kmeans++ paper) cited the incorrect hardness result till then.
Even when I read a paper, I find some minor mistakes. However, they are easily fixable and do not change the main result very much.
I want to ask if there had been any fundamental or highly cited paper, which was later found to be incorrect, and due to which the entire understanding of the field changed.
Edit: After reading some of the answers, I want to point out another issue that, why these incorrect papers are not updated after being pointed out wrong. I mean some kind of notice must be provided by the governing body who shares the link. In my case (for the example I gave above), it took me two years to figure out why people in the year 2002 were designing the PTAS for the k-means problem(for fixed k) if the result of hardness came later in the year 2009. It can be quite frustrating for a person who is not familiar with that field.