Compared to the typical American, Americans with disabilities are about twice as likely to be unemployed and they are two-thirds less likely to even enter the workforce. These facts alone should give policymakers pause. Eighty percent of people with disabilities who are not employed report that the obstacle is the disability itself.

For decades, the story has been a frustrating one: lower employment rates, persistent discrimination, and heavy reliance on disability insurance. The Americans with Disabilities Act was meant to change that. But research suggests it had an unintended consequence: decreased employment among the disabled, possibly by discouraging hiring by raising perceived costs and legal risks for employers.

Now comes artificial intelligence—a technology that could either deepen these challenges or help dismantle them.

On its face, AI offers remarkable promise. For workers with visual impairments, tools can describe surroundings, read documents and identify objects in real time. For those with hearing impairments, real-time captioning and translation are making meetings and digital communication far more accessible. Workers with mobility limitations can use voice-activated systems to control their environments, while AI-powered prosthetics and assistive devices improve independence. And for people with cognitive or learning differences, AI can summarize complex information, provide prompts and tailor learning to individual needs.

In short, AI can reduce the functional barriers that have long limited people with disabilities’ participation in the workplace.

But the same technology also introduces new risks. Hiring algorithms may screen out candidates who don’t conform to expected speech patterns, facial expressions or behavioral cues. Voice-activated systems might exclude those with speech impairments. And when designers fail to include people with disabilities in building AI tools, accessibility becomes an afterthought rather than a feature.

The result is a paradox: AI can both level the playing field and tilt it further.

Early research reflects that tension. One recent study finds that AI reduces employment for people with disabilities, as automation and skill mismatches take hold. Over time, however, those effects soften and can reverse—especially as workers gain skills and as AI complements rather than replaces human labor. Education plays a critical role: disabled individuals with higher levels of training are far better positioned to benefit from AI-driven workplaces.

There are also important differences across groups. Men with disabilities are more concentrated in routine or automation-prone jobs, causing them to face larger short-term employment losses. Women with disabilities experience smaller—though still meaningful—effects.

A study out of the University of Toronto offers a deep dive into how AI can greatly benefit the disabled and their employers when companies adopt adaptive tools. The study is of a large Chinese food delivery platform, much like Uber Eats, that completes hundreds of millions of deliveries per month. The platform is one of the largest food delivery platforms in the country and employs large numbers of people with hearing impairments, under an open hiring policy.

Before the company introduced AI tools, disabled workers were less efficient than their hearing counterparts—particularly in tasks requiring communication with customers—being 8 percent more likely to have late deliveries and 37 percent more likely to receive a bad customer rating. They also worked 7 percent more hours per week.

Then the company introduced an AI-powered communication system that allowed hearing-impaired workers to interact with customers, restaurant workers, and others using natural-sounding AI voices.

The results were striking. Workers became faster. Late deliveries and negative customer ratings declined. Productivity increased. And the wage gap between the hearing impaired and their peers shrank by more than half, from 8 percent to 3 percent. These workers became even more valuable to the firm: Incomes increased 11 percent.

AI didn’t replace workers with disabilities. It made them better and more valuable.

That distinction matters for policy.

The lesson is not that AI will automatically improve employment outcomes for people with disabilities. It is that outcomes depend on how the technology is used, deployed and governed. Used the wrong way, AI can exclude. When it complements human abilities—particularly by addressing specific barriers—it can unlock human potential.