Wiles was weird at the time. Math is done in public, he just really wanted to be the one to have the final valid proof of Fermat's Last Theorem.

Saying you’re working on FLT in the late 20th century is a good way to make everyone think you’ve gone nuts and are wasting your time. It makes some sense to make sure you have something solid before you mention it.

Technically he was working on the Taniyama conjecture (now known as the modularity theorem), not on Fermat's last theorem.

or conversely, if they believe you, they could try to sniff your approach and take the credit for it. given the fame of FLT I dont blame him for doing it undercover.

Graph isomorphism isn't that important practically (subgraph isomorphism is, but it's NP-complete and, IIRC, APX-hard as well) and the hard graph cases don't come up much in practice (despite being a relatively large percentage of random graphs, AFAIK). I think it reflects a difference in how mathematics is done; the quasipolynomial result was first presented at a conference, not in a paper, which definitely wouldn't have been the case 20 years ago.

Wiles' original FLT result was first presented at a workshop in Cambridge, although he didn't explicitly state that he had succeeded (sans the gap) in proving FLT, he left that for the audience to infer.

And that doesn't even make it unusual. Most mathematical results are first presented at conferences or workshops, sometimes long before publication.

I was around at the time. I remember the buzz after day one of his two day workshop. Everyone knew where he was going.