I have seen a few proofs of the claim. One that abuses the Riemann zeta function [itex]\zeta (s) := \sum_{k=1}^\infty \frac{1}{k^s}[/itex]. This is valid only when [itex]Re(s)>1[/itex]. and they conclude result by
[tex]
\zeta (s) = 2^s \pi ^{s-1}\sin\left (\frac{s\pi}{2}\right )\Gamma (1-s)\zeta (1-s)\vert _{s=-1} = -\frac{1}{12}[/tex] which, indeed, is true, but its association to the series representation is invalid. Then there is the geometric series "proof" where they say [itex]\sum_{k=0}^\infty x^k = \frac{1}{1-x}[/itex] and they do mention that it is valid when [itex]x<1[/itex], but then conviniently set [itex]x=-1[/itex] and conclude their result. In reality, the sum is valid when [itex]|x|<1[/itex] i.e they cheated. It's known as ex falso quodlibet. Proceed under false assumptions and conclude that the Sun is smaller than the Earth.. hooray. Not quite -.-
If we involve analytic continuation then we are no longer talking about the sum in the traditional sense. This is the part people get confused, because nobody explicitly states that we aren't actually talking about the sum in the sense of addition. It's more like, we assign a value to this series and prove that the operation is valid. But that's not exciting, is it?
I wouldn't be studying pure mathematics if the principles of logic were so easily defied without consequence. The heretics may preach what they wish, at that point, what they dabble in is no longer backed by logic, therefore not mathematics.