At least with overlays, I have some agency. I can turn them off, block the script. I can see what was made worse.

With Accessibility Firewall, it seems I cannot (based on its marketing). Cannot stop it, cannot see what changed, cannot undo it.

Overlay claims with worse outcomes.

Accessibility Firewall ‘fixed’ one of the 133 issues axe DevTools identified on Vercel’s home page. By adding a useless `aria-label` to the logo link.

Meanwhile, it introduced more WCAG issues, while also removing subnav, login, sign in, “Ask AI”, etc.

Impressive.

Using Vercel, Ashlee found the Accessibility Firewall introduced *more* errors:
https://bsky.app/profile/ashleemboyer.com/post/3mkvg3kfjq22g

Curious, I checked.

When faced with the simplest error (a disclosure widget), its solution was to remove it — and site navigation for all mobile / narrow viewport users.

Cool.

First post in thread, I found Accessibility Firewall gave itself a 73% ‘score’.

Ashlee checked under 20 minutes later and it jumped to 78%:
https://bsky.app/profile/ashleemboyer.com/post/3mkt2ehq24k24

Kevin a couple hours later and it was 83%:
https://bsky.app/profile/kevinpowell.co/post/3mktc6hyqgs2e

Either they’re live-fixing or the scores are bunk.

Seems a bit… random?

Tired: #overlay companies lying about using ‘AI’ for WCAG remediation (accessiBe’s FTC fine, my UserWay tests, etc.).

Wired: ‘AI’ companies lying about not being overlays (sans opt-out).

No accessibility knowledge needed to evaluate these. Basic HTML and reading skills uncovered this.

Firewall claims:
• “Your website is remediated before it reaches the browser.”
• “The user receives mathematically compliant HTML. Fully accessible…”

Webcraft claims:
• “…instantly ensuring compliance…”
• “Our full remediation engine processes all 86 WCAG 2.2 criteria…”
• “Certified accessibility experts review the results to ensure legal defensibility.”
• “Our engine patches them at the edge, ensuring compliance and integrity.”
• “…enforcing Section 508 compliance…”

I made archive pages of the Firewall product:
• https://web.archive.org/web/20260501233848/https://accessibilityfirewall.com/
• https://archive.is/wcHGD

And Webcraft itself:
• https://web.archive.org/web/20260501234507/https://webcraft.inc/
• https://archive.is/KkZvg

These may be useful in future discovery during false advertising litigation. Because they make bold claims (in order):

I needed to know.

https://vpatcode-worker.webcraft-stephanie-fenton.workers.dev/proxy?url=https%3A%2F%2Fadrianroselli.com%2F

It asserts one issue and a fix, and I think it’s wrong. Not just the issue but also the SC.

Doing the math, there are 14 issues remaining. Though, maybe those are the “13 human-in-the-loop recommended”?

So the numbers feel made up, the remediation feels sketchy, the language feels off, and I feel this might be less than advertised.

But at least the Terms & Privacy policies are brief enough and offer sufficient recourse to make me walk away.

• https://webcraft.inc/terms
• https://webcraft.inc/privacy

I pulled the nodes out for the non-dismissable banner and then grabbed the before and after HTML. An online diff show the changes, or remediations, were minimal.

Moved the `<base>`, added whitespace, 3 contrast ‘fixes’, and suppressing Cloudflare cookie consents and errors.

Nice outlines.

The banner introduces WCAG violations, per axe. Though a user should never see the banner.

But I wondered what went from an opaque score of 73% to 100%, or which 3–6 issues were fixed.

Sadly, the banner dismiss button didn’t work. Either way, I saw no change (contrast was false positive).

When I visited the ‘remediated’ version of the firewall product in two browsers, I got a different number of auto-fixed items.

Maybe you’ll get a different count?
https://vpatcode-worker.webcraft-stephanie-fenton.workers.dev/proxy?url=https%3A%2F%2Faccessibilityfirewall.com%2F

Pointing a tool at the vendor’s site is often a good test.