A hiring AI scored "I have autism" more negatively than "I am a bank robber."
That's not a hypothetical. Researchers at Duke found that large language models consistently associate neurodivergent terms with negative meaning. And these same models are powering the tools that decide who gets a job interview.
This is the reality the ACLU exposed when they filed an FTC complaint against Aon Consulting last year.
The short version: Aon sold personality tests and video interview tools marketed as "bias free." But the ACLU found that the traits being measured — things like "liveliness," "positivity," and "emotional awareness" — map almost directly onto clinical diagnostic criteria for autism.
Think about that for a second.
An algorithm trained to favor "socially bold" and "outgoing" candidates will mathematically screen out autistic applicants. Not because they can't do the job. Because they don't perform neurotypicality the way the machine expects.
Our team has been deep in this research, and here's what keeps coming up: the problem isn't just one company or one tool. It's an entire generation of hiring tech built on a hidden assumption that "good employee" equals "neurotypical employee."
The fix isn't surface-level auditing. It requires rethinking how these systems measure people in the first place — separating actual job performance from personality theater.
We wrote up the full technical breakdown, including what enterprises can do right now to close these gaps. Link is in the comments.
Honest question for this community: if you found out an AI rejected your application based on personality traits tied to a disability you might not even know you have — would you ever trust automated hiring again?
#AlgorithmicBias #AIGovernance #Neurodiversity