Reppert on the Dragon, and You Can't Always Believe What You Read
Now, to be fair, theory develops, and what really looked safe and sound at one time may turn out otherwise. Chess is to we limited and fallible human beings as good as infinite in its riches, so we have to discover things a bit at a time. Many - hopefully most theoretical errors in books reflect nothing more than the limits of our abilities even after we've tried our best.
But not all, unfortunately. Here's one particularly odious example, from Alexei Shirov's generally magnificent book Fire on Board (note: many of the annotations in the book are based on notes first given in various publications shortly after the games were played). In game 26, Shirov-Kozul, Biel 1991, Black played 15...b5. Shirov gives this move the dubious sign ('?!') and writes this: "Preferable was 15...Bb5 16.Bxb5 Qxb5 17.Kf2 Rxc1 18.Qxc1 Nc6 19.Qb1 Qa6 20.hxg6 hxg6 21.e5 Rd8 22.Qe4 Qxa2 23.Qh4 Kf8, as in Shirov-I. Gurevich, World Junior Championship, Santiago 1990)."
However, just a few weeks later in the Lloyds Bank tournament in London, Shirov-Ernst continued with the same first 21 moves as Shirov-Gurevich, but then White sprang 22.a4!, winning five moves later (though Black could have reached an acceptable but inferior position with best play).
Did Shirov just discover this improvement between the time of his published notes to the first game and his playing the second one? Unfortunately not. In Shirov's comments for the book, he admits that the note with his suggested "improvement" on move 15 of game 26 was "just a trick." Professionals are probably all fully aware that this sort of thing takes place, but amateurs, beware!