I've never heard of an "autobiographical" implicit association test, but personally I'd be skeptical of using this for lie detection.
All implicit association tests are essentially the modern version re-tooled version of Freud's free association method which, in itself, took a lot of time and many sessions to be effective. The instant gratification of the IAT doesn't allow for such basic considerations as short term variance in the subject's state. Associations will be stronger or weaker day to day, depending on what the subject has accessible and primed at the moment.
Moreover, it doesn't consider individual differences in overall accessibility due to who they are and what they do... just because someone easily groups "axe" "bloodstain" and "kill" doesn't mean that they're an axe murderer -- they could be a horror fan.
From my uneducated psychology geek perspective -- this whole thing is a really bad idea.
RE: Polygraph Alternatives: Could Reaction Time Catch a Liar?