Ship It Fast or Build It for Everyone? India's AI Rush and Its Disabled Citizens

Ship It Fast or Build It for Everyone?

India wants AI everywhere and it wants it quickly. Models are moving into government services, recruitment and hospitals and the Centre is courting $200 billion in AI investment over the next two years. The question this raises for disabled citizens is not whether AI helps them - it plainly does - but when accessibility should be required: before a system is deployed or after it is already in use.

The case for moving fast The strongest argument for speed comes from disabled users themselves.

Among 2,462 people surveyed on NClude, a platform that uses AI to help disabled users complete job applications and government forms on inaccessible websites, 1,313 finished a task they could not do before.

• A blind citizen can now read a document or get through a government form without waiting for a sighted helper.
• Every month a useful tool is held back for compliance review is a month of that independence lost.
• India still has not finished framing binding accessibility standards for ordinary websites - so tying AI deployment to standards that do not yet exist could freeze the tools people are already relying on. This is not a weak argument. It is the argument disabled users make when they say: do not take away what works.

The case for building it right first The same data reveals the flaw.

Of those 1,313 successes, only 543 came through AI alone; the rest needed a staff member to step in. Without that human fallback, the user is left to improvise.

• The AccessEval benchmark found 21 language models became more error-prone and more likely to stereotype when disability entered the question.
• The CLIP image model performed 15 percentage points worse on photographs taken by blind and low-vision users.
• White canes and Braille displays appeared up to 17 times less often in training data. And India already knows what "fix it later" looks like. The RPwD Act has required accessible digital services since 2019. In Rajive Raturi (2024), the Supreme Court called the rules toothless and ordered binding standards within three months. Nearly two years later, the petitioners are back in court and 155 establishments, including ministries, have been penalised for inaccessible websites.

Where

TAN stands TAN's position is that accessibility testing must be mandatory at the procurement stage - whenever the government buys or deploys an AI system for public use. This does not stop private developers or slow tools like NClude.

It uses the one lever the state fully controls: its own purchasing. Any AI system a ministry deploys should pass a disability-bias test and include a human fallback. Because Raturi binds the state under Article 21, the government cannot claim speed as a reason to deploy systems that predictably exclude the citizens it serves.

The pace of AI does not have to slow. What must change is who checks the system before it reaches the public counter.