- The ACLU proved that a major hiring AI asks questions nearly identical to autism diagnostic criteria.
It's not screening for job fit. It's screening for neurotypicality.
Most enterprises don't even know their tools do this. 🧵 - In May 2024, the ACLU filed an FTC complaint against Aon Consulting.
The allegation: Aon marketed its AI hiring tools as "bias-free" while those tools likely screen out candidates based on race and disability. - Aon's ADEPT-15 test scores candidates on traits like "liveliness," "positivity," and "emotional awareness."
These aren't neutral personality traits. They're proxies for neurotypical social performance. - Map ADEPT-15 constructs against the Autism Spectrum Quotient and DSM-5 criteria.
The overlap is damning.
"Reserved" = penalized. "Insulated" = penalized. "Consistent" = penalized.
All common autistic traits. None are job deficiencies. - It gets worse. Aon's vidAssess-AI uses NLP trained on neurotypical speech patterns.
Flat intonation? "Lacks confidence." Atypical pauses? "Disorganized."
The AI pathologizes neurodivergent communication styles as performance failures. - Duke research found LLMs rate "I have autism" more negatively than "I am a bank robber."
When these models power hiring tools, they embed that bias into every screening decision — invisibly.
The FTC isn't treating this as an ethics debate anymore. It's enforcement. - "Operation AI Comply" means every "bias-free" claim now requires empirical proof. No substantiation = deception under Section 5.
The EEOC has made it clear: employers own the discrimination their vendor tools create. - "We didn't build the AI" is not a defense. You deployed it. You're liable.
The fix isn't better wrappers. It's causal AI. - We use Causal Representation Learning to mathematically isolate decision logic from protected attributes. Not correlation cleanup — structural guarantees of counterfactual fairness.
Every enterprise using AI hiring tools right now faces a choice: - Audit deeply, or wait for the complaint.
The Aon case is the first. It won't be the last.
Should companies be required to disclose when their hiring assessments overlap with clinical diagnostic criteria — or is that an unreasonable burden? #AIGovernance #AlgorithmicBias - We wrote the full technical breakdown — architecture flaws, regulatory shifts, and the deep AI framework that actually fixes this: https://veriprajna.com/whitepapers/algorithmic-ableism-crisis-deep-ai-governance-enterprise-hiring