# uBlock Origin fixed a one-click path to running code on other sites

**Summary:** uBlock Origin's element picker could let hostile pages add filters that run code on other sites. Version 1.75.4 fixes the flaw; 1.76.0 adds hardening.

- Canonical: https://markhuang.ai/news/ublock-origin-element-picker-uxss
- Language: en
- Author: [Mark Huang](https://markhuang.ai/about)
- Published: 2026-11-16
- Section: News
- Tags: uBlock Origin, Firefox, Browser Extensions, Web Security, Clickjacking
- Source: [Smitty van Bodegom](https://iter.ca/post/ubo-uxss/)
- License: https://creativecommons.org/licenses/by-nc/4.0/

---

![A translucent pane conceals a mouse pointer between two paper windows, with a red ribbon crossing through a shield](https://cdn.markhuang.ai/news/ublock-origin-element-picker-uxss/hero.webp)

*Conceptual illustration: a concealed interaction with an extension can carry authority from one website to another. Generated with imagegen; this is not a screenshot of the exploit.*

A click on a hostile webpage could have made uBlock Origin run JavaScript on another website. The route went through its element picker, the tool that lets you select an unwanted part of a page and create a rule to hide it. An attacker could make that tool save a rule with much more power than hiding an annoying box.

In his [November 15, 2026 disclosure](https://iter.ca/post/ubo-uxss/), researcher Smitty van Bodegom describes the flaw in the Firefox version of uBlock Origin built on Manifest V2, an extension platform. The [project's advisory](https://github.com/gorhill/uBlock/security/advisories/GHSA-cmqw-9r2w-6wv2) lists versions 1.75.2 and earlier as affected and 1.75.4 as patched. Additional hardening followed in 1.76.0.

The picker could create trusted filters that inject JavaScript, the code webpages run. By getting the extension to save one for a target domain, a malicious page could have its code run in that other site's context. That reach explains the report's name for the flaw: universal XSS, or universal cross-site scripting.

## The picker had authority the webpage should not have

Normally, browsers limit what one website can do to another. [MDN's explanation of the same-origin policy](https://developer.mozilla.org/en-US/docs/Web/Security/Same-origin_policy) uses a familiar example: a malicious site should not be able to read a signed-in user's webmail. The concern here was that the extension could become a route around that separation.

uBlock Origin's picker lives in an extension page embedded inside the webpage as an iframe, a small document within another document. It communicates with the extension's background code to save filters. According to van Bodegom, a hostile page could embed its own copy of the picker and send messages that the picker accepted.

The attacker first needed the picker's extension address, which includes a random identifier assigned to that installed extension. The report shows how a filter that substitutes an extension-provided script for an advertising request could expose that identifier through a web request the page could observe. That made the picker reachable.

The attacker could then send a crafted message. A field meant to say which type of request a rule should match could contain a newline and a second rule. The picker normally restricted the kinds of filters a user could enter, but this route let the attacker smuggle in a trusted filter capable of injecting code into another site.

The sequence in the report, including its proposed concealed-click step, was:

```mermaid
flowchart TD
  A["Hostile page learns the extension address"] --> B["Page embeds the element picker and sends it messages"]
  B --> C["A crafted field includes a rule for another website"]
  C --> D["User clicks Create, potentially concealed by the page"]
  D --> E["Extension saves a trusted filter"]
  E --> F["Filter runs attacker-controlled JavaScript on the target site"]
```

## One click, with an important demonstration limit

The click was still necessary. The researcher explains that a malicious page could make the embedded picker invisible or cover it with other content, so an apparently ordinary click would press its Create button. This is [clickjacking](https://developer.mozilla.org/en-US/docs/Web/Security/Attacks/Clickjacking): the user physically clicks a sensitive control without realizing which action it performs.

Van Bodegom explicitly says he did not implement that concealment step. The email reproduced in the advisory describes a proof of concept with the button visible. The demonstration showed the filter-injection path; hiding the button was the author's proposed way to disguise the required click. I have not reproduced either.

Changing custom filters later would offer limited protection, the report explains. Editing custom filters causes uBlock Origin to recompile them and clear their trusted status. But an attacker could open the target page immediately after adding the rule, before the user made such an edit.

The author mentions a related, less problematic issue in uBO Lite, which uses Manifest V3. This disclosure's detailed attack concerns Firefox's Manifest V2 version; it does not establish an identical outcome across every browser or extension edition.

## Why the fix went beyond hiding the address

The advisory records an initial change in 1.75.2 to avoid web-request methods that the page could tamper with. The researcher then reported another way to expose the extension identifier. Blocking one disclosure route had left another available.

The [secret-token change](https://github.com/gorhill/uBlock/commit/9859477ebc647352282ddd71b5013ce4764493ed), released in 1.75.4 on November 3, required the picker to verify a secret supplied by the extension's own code before starting. Version 1.76.0, released November 8, added validation of the filter about to be created and temporarily prevented filters using the extension-resource response mechanism involved in the report.

For extension developers, I think this is the most useful part of the fix. Knowing where a privileged interface lives should not authorize a webpage to operate it. Checking the caller and validating the rule it actually submits address separate parts of this failure. A button click alone cannot establish informed permission when the surrounding page controls what the user sees.

## AI helped find it; verification made it actionable

Van Bodegom credits GPT-5.6 Sol with most of the work of finding the issue after he suggested looking for filter injection. He says he checked and verified the proposed issue, while Claude wrote most of the proof-of-concept code. His account includes both a human hypothesis and verification of the result.

That verification work also appears in the [review workload around AI-generated security reports](https://markhuang.ai/news/anthropic-oss-scanner-curl-review-queue): a candidate finding still needs someone to establish what fails and under which conditions. Here, the advisory lets readers connect the reported failure to specific fixes.

A commenter in the [Hacker News discussion](https://news.ycombinator.com/item?id=50513160) raises the extension's broad ability to inspect and modify pages. For users, the immediate decision is concrete: check the installed uBlock Origin version and update to the current stable release. Keep the milestones straight: 1.75.4 is the advisory's patched version; 1.76.0 adds further hardening. For builders, the repair suggests a concrete check: authenticate messages to powerful embedded tools, and validate the final action they authorize.
