3 ms·
I was kind of excited reading the headline, but this seems mostly to be a pretty straight forward application of existing homoglyph attacks that have been commo
by bawolff 5y ago
I was kind of excited reading the headline, but this seems mostly to be a pretty straight forward application of existing homoglyph attacks that have been common for decades at this point (albeit maybe not common against these types of programs previously).
I thought it was going to be adding imperceptible changes that allow you to choose the interpretation (sort of like the visual adversarial examples) not just substitute a homoglyph and the binary classifier no longer works.
Don't get me wrong, this is still important work, i just got a bit too excited.