I definitely remember rumblings about "Branch prediction runs code outside normal execution? There's got to be a security hole there somewhere." That sentiment was common enough that it's certainly not hard to imagine someone sketching the shape of an actual attack with it before the detailed proof came down the line.
