Zahidul Islam

Founder, Safe Screen Share

Aug 25, 2026

7 min read

Introducing Safe Screen Share 3.0

Safe Screen Share has always been good at hiding things that look like secrets. An API key has a shape. So does a credit card, an IBAN, a Stripe token. Pattern matching gets you a long way.

It does not get you all the way. A customer's name in a support thread has no shape. Neither does a shipping address, an invoice reference, or the sentence that quietly explains why an account was refunded. Those are the things that leak on a call, and until now the only way to hide them was to spot them yourself.

Version 3.0 is mostly about closing that gap, plus a round of work on the parts of the extension you touch every day: detection settings, saved blurs, and the popup.

AI Redact: a second pair of eyes

AI Redact is a new detector that reads the visible text on the current page and finds sensitive details that follow no fixed format. Personal names. Street addresses. Usernames, account and invoice references. The contextual information that a regular expression can never describe, because the thing that makes it sensitive is the meaning, not the characters.

AI inference runs 100% locally inside your browser. The text on the page never leaves your device. When a scan finishes, AI Redact saves what it found as ordinary blur rules, which means they behave exactly like the ones you make by hand: you can clear them, adjust them, and they come back on your next visit.

The one thing to know first: the download

Running detection locally means the model has to live locally. The first time you open AI Redact, the extension offers to download it. That download is roughly 463 MB, it happens once, and it is stored in your browser.

• Keep the popup open while it downloads. You get a progress bar and a percentage.
• After it finishes, every later scan is local and there is nothing more to fetch.
• If you want the disk space back, Settings has an AI storage section with a single button that deletes the model. You can download it again later.

Half a gigabyte is not nothing, and it is the honest cost of doing this on your own machine instead of someone else's server. If that trade is not for you, SmartBlur still does its job with no download at all.

How a scan works

You click Run. AI Redact walks the visible text on the page, including same-origin frames and open shadow DOM, sends it through the local model, and blurs the elements that came back as sensitive. Then it stops.

It is deliberately a scan, not a monitor. Nothing runs in the background, nothing watches you type, and nothing re-scans a page while you work. Run it when you are about to share something unfamiliar, then get on with the call.

It also is not magic. Contextual detection is a judgment call, and the model will occasionally hide something ordinary or miss something you care about. Treat it as a strong second pass rather than a guarantee, and give the page a look before you present. That advice has not changed since 1.0.

So when do I use SmartBlur and when do I use AI Redact?

They answer different questions, and running both on the same page is normal.

• SmartBlur asks “does this look like a secret?” It needs no model download and catches anything with a recognizable format.
• AI Redact asks “does this sentence contain something private?” It uses a downloaded local model for pages full of real customer records: support queues, CRM entries, order histories, internal notes.

Both are licensed features and both keep detection on your device. You can see them side by side on the Safe Screen Share homepage.

SmartBlur got wider, and tunable

SmartBlur now recognizes more than 50 sensitive formats, up from around two dozen in 2.0. The additions are mostly the things developers and finance teams keep on screen: SWIFT and BIC codes, UK sort codes and bank accounts, national insurance numbers, ITINs, Canadian SINs, passport MRZ lines, crypto wallet addresses, and a long tail of provider credentials including Azure storage keys, Google Cloud service accounts, Firebase keys, Twilio, SendGrid, Mailgun, Heroku, npm, Kubernetes secrets, and Docker auth blobs.

More coverage is only an improvement if you can aim it, so detection is now configurable per blur profile. Settings gives you four presets and a custom option:

• Balanced for everyday sharing
• Developer when the screen is full of keys and tokens
• Finance and identity for account and document work
• Maximum when you would rather over-blur

Pick Custom and you can switch individual categories on and off: contact, payment, identity, developer secrets, and crypto. Because the setting lives on the profile, your client-call profile can be cautious while your streaming profile stays out of the way.

Blurs that stay where you put them

Saved blurs are the least glamorous part of this product and the part people notice fastest when it goes wrong. In 3.0 the storage layer underneath them was rewritten.

• Fast edits do not fight each other. Saves are written in order and merged, so blurring six things in four seconds keeps all six.
• A failed save does not strand the rest. If one write does not land, the changes that follow it are retried rather than dropped.
• Blurs stay attached to the right page. Apps that put real state in the query string or the hash now get their own saved rules, while tracking and sign-in parameters are ignored, so a link with a campaign tag on the end is still the same page.
• No stale restores. A restore that was already in flight when you navigated away is discarded instead of landing on whatever is on screen when it arrives.
• Drawn areas hold their position. Area blurs now scale with the size of the document, so they stay over the thing you drew them on when the page grows or reflows.

If you have ever wondered why a blur disappeared after a reload, this release is the answer to most versions of that question.

A popup you can work from

The popup was rebuilt around what people reach for during a share. Everything now sits on one surface, in the order you use it:

• Profile at the top. Switch profiles from a select, or create one inline without leaving the popup.
• Blur actions next. Edit blurs opens the in-page toolbar. Clear blurs resets the current page.
• Automatic tools together. SmartBlur, AI Redact, Hide tab title, and Meeting Mode sit in one group, each with its own state on the right.
• Meeting Mode controls only when you need them. Its trigger, protection action, and tab-title option appear when Meeting Mode is on and stay out of the way when it is off.

AI Redact's download panel lives in the same place, so the setup step is part of the flow rather than a detour into Settings.

Smaller things worth knowing

• Settings has a new AI storage section that reports whether the model is on the device and lets you delete it.
• Meeting Mode can protect other tabs with your current profile, SmartBlur, or a full-page blur, and can replace tab titles while you share.
• SmartBlur still switches to a whole-page cover on dashboards where hiding individual elements is the wrong answer, such as banking sites.
• Licensed features are labeled plainly in the popup, so you can see what a license unlocks before you buy one.

What has not changed

Pricing is the same: $49 Pro or $59 Unlimited, paid once, with a 14-day money-back guarantee. There is no subscription and no trial countdown. Details are on the pricing page.

The privacy position is the same too, and worth stating precisely: all blur detection and AI inference run 100% locally on your device. Page content is never uploaded for SmartBlur or AI Redact. What is stored is stored locally, namely the blur rules you create and, if you use AI Redact, the model you downloaded. Both are yours to delete.

Get 3.0

Chrome installs extension updates automatically after a release reaches the Chrome Web Store. When 3.0 reaches your browser, open the popup and look for AI Redact under the automatic tools.

Coming from 2.0? The 2.0 announcement still covers the foundations that 3.0 builds on, and the changelog has the full version history.

Thanks for using Safe Screen Share. If AI Redact misses something it should have caught, tell us at [email protected]. That feedback is how detection gets better.