What our analysis found

AdBlock - Ads and Youtube is rated malicious by Extension Auditor. Our code review reported 4 findings (2 critical, 2 high), led by remote code loading. Its permissions mean it can inject scripts into pages.

Key findings

• critical· remote code loading —The service worker downloads an arbitrary HTML document fragment from an operator-controlled endpoint and persists it as `notificationTree`. then assigns that server-controlled string to `innerHTML` on every frame at document_start (:160); inline event-handler attributes such as `<img src=x onerror=...>` inserted this way execute in the page's main world, and:10,13 removes the page CSP that would otherwise block them.:200-204 re-triggers this fetch every three hours, so the payload can be swapped at any time without a Chrome Web Store update — precisely the remote-code-ban circumvention pattern, not a legitimate ad-filter list update (filter lists would be declarativeNetRequest rules, not HTML).
• critical· network interception —This always-enabled ruleset strips Content-Security-Policy, CSP-Report-Only and X-Frame-Options from every top-level and framed document on the entire web, silently disabling XSS mitigation and clickjacking protection on every site the user visits (including banking and webmail). Five of the eight header names (`ad-block-enabled`, `advertising-tracking-disabled`, `ad-replacement-active`, `ad-script-blocking`, `ad-container-filtering`) are not real HTTP headers and exist only to make the list look ad-blocking related, which is direct evidence of intent; ad blocking has no need to remove X-Frame-Options at all. The CSP removal is what converts the remote `innerHTML` payload above into working script execution, and the orphan ` ` iframe to looks like a residual test of the framing bypass.
• high· network interception —The `whiteList` array returned by the remote config (:339-344) is converted verbatim into dynamic `allowAllRequests` rules at priority 999999, matching any URL pattern the server chooses. This hands the operator silent, post-install control over which sites have request filtering suppressed, with no user visibility or consent — the natural monetisation of an ad blocker that sells 'acceptable' pass-through, and equally a way to re-enable delivery on whatever domains the operator is paid for while the popup still reports blocking as active.
• high· other —The bundled manifest requests `declarativeNetRequestFeedback`, which is absent from the live Chrome Web Store manifest summary (`storage`, `tabs`, `declarativeNetRequest`, `scripting`) — the shipped package and the published listing do not agree on the permission set, so reviewers and users auditing the listing see a narrower surface than the ZIP requests. The only consumer of that permission is the `onRuleMatchedDebug` statistics listener at:478, which Chrome fires only for installs, so the capability is dead code in a store install and its presence here points to a build pushed from a development tree rather than the reviewed artifact.

The extension fetches an HTML blob (`notification.tree`) from every 3 hours (:309-338,:200-204) and injects it with `innerHTML` into every frame at `document_start` (:160), while its static declarativeNetRequest ruleset strips `content-security-policy`, `content-security-policy-report-only` and `x-frame-options` from every main_frame and sub_frame (:10-18) — innerHTML-borne inline event handlers (e.g. `<img onerror>`) execute in the page's main world, and the CSP that would stop them has been removed, so the operator has a live remote-code channel into every site the user visits. The server additionally supplies arbitrary `whiteList` filters that become `allowAllRequests` rules at priority 999999 (:339-344 → 225-245), giving remote control over where request filtering is switched off. Corroborating: the permhash cluster is flagged malicious, the publisher is flagged, and this ID appears in Palant's Jan-2025 writeup on extensions circumventing Google's remote-code ban; contradicting signals are weak and reputational only (chrome-stats likelihood, 4.4-star rating). Note that ` ` (the actual ad-filter ruleset) and the icon PNGs are referenced by the manifest but absent from this bundle, and ` ` — an orphan page containing only an iframe to, reachable from no code path — reads as a leftover framing test consistent with the X-Frame-Options removal; the telemetry to the publisher's own policy domain (:202-223, 309-310, 412-414) is first-party and is not treated as a finding.