Hold the Horses on AI Automated Scan Remediation
A new sentiment is floating that AI will be able to automatically fix any issue flagged by a scan. While this will eventually happen, we’re not as close as you might be led to believe.
One big reason why is AI makes mistakes and hallucinates in execution. Even in scenarios that are completely logical, even the best AI models continually arrive at the wrong conclusion.
When it comes to AI fixing issues flagged by AXE, WAVE, or another scan, this means you could have an agent that autonomously begins to fix issues, but then makes any number of errors in execution.
We’ve seen this over and over in all types of scenarios. It’s critical to know that artificial intelligence is very fallible and can and does make very elementary mistakes.
And you definitely don’t want AI directly fixing code without oversight because there will be mistakes and omissions made and those mistakes and omissions will completely undo all of the efficiency that was supposed to be gained in the first place.
Not only this, but we can easily see a scenario where so much goes wrong in a staging environment that the best option is to completely reset and go back to the beginning.
The other problem is that errors in accessibility fixes are often unknown to the non-experts reviewing them. A fix looks correct in the code. The scan passes it. But a screen reader user hits a wall that wasn’t there before. Nobody catches it.
This is what makes unsupervised AI remediation a beta experiment for a while. It doesn’t just leave things unfixed — it can create new problems (accessibility and otherwise) and it produces a record that says things were fixed. And real accessibility issues remain underneath a clean scan. That gap between what the scan shows and what users actually experience is where the problem lives.
