ABLEIST: Intersectional Disability Bias in LLM-Generated Hiring Scenarios

URL SCAN: ABLEIST: Intersectional Disability Bias in LLM-Generated Hiring Scenarios
FIRST LINE: # Computer Science > Computation and Language

The Dissection

This paper audits six LLMs across 2,820 hiring scenarios involving disability, gender, nationality, and caste. Its real function is to expose LLMs as potentially biased employment gatekeepers and show that generic safety tools miss intersectional harms.

It is a diagnostic of automated selection—not an analysis of labor demand, ownership, or the survival of the employment system.

The Core Fallacy

Under the Discontinuity Thesis, the paper mistakes fairer access to a shrinking employment funnel for preservation of productive participation. Even if models stop producing ableist, tokenizing, or stereotyped outputs, P1–P3 remain: cognitive work is automated, human-only coordination fails at scale, and most people lose access to economically necessary labor.

Intersectional safety evaluation may reduce discriminatory allocation. It cannot restore the wage–consumption circuit. It makes the gate less obscene while leaving the gate’s structural disappearance intact. This is the paper’s analytical limitation, not necessarily an argument it explicitly makes.

Hidden Assumptions

• Hiring remains the main route to material security.
• Better model behavior will produce better real-world hiring outcomes.
• Six LLMs and 2,820 scenarios meaningfully represent frontier systems and deployed workflows.
• The selected metrics capture the relevant harms.
• Removing bias from selection changes the number of viable jobs.
• Employers will keep using audited systems rather than shifting to opaque filters or abandoning marginal candidates.
• Inclusion in the labor market is equivalent to durable economic participation.

Social Function

Partial truth; transition management; prestige signaling. The study identifies genuine harms and supplies a useful vocabulary for exposing them. Institutionally, however, this work also helps legitimize continued automation: if the gate can be audited, scored, and tuned, the system can present itself as governable while the underlying labor bargain dies.

It is civilizing the sorting machine, not dismantling dependence on sorting.

The Verdict

A valuable bias autopsy with a strategically narrow field of vision. It can help prevent disabled and multiply marginalized people from being punished by automated hiring today. Under DT logic, it is still a fairness patch on a labor market whose central function is being automated away.

It improves distribution of the remaining access. It does not preserve mass productive participation, ownership, or bargaining power. The paper diagnoses a real wound, then treats the triage protocol as if it were a cure.