AdBlock — block ads across the web

AI Security Review

Evidence context: threat category none; evidence quality moderate.

AdBlock 6.49.0 ships as a Chrome extension with roughly 65 million users and a developer contact of [email protected], the address the AdBlock team has used publicly for years. The manifest we unpacked declares an empty permission list and an empty host permission list, and the extraction resolved no external network endpoints at all. That shape matches a content blocker that keeps its filter rules in static files rather than in code that phones home.

The one high-severity item is YARA--supply_chain_sourcemap_appended_iife on /abp-background.js.map line 1. A .map file is a compiler artifact that lets devtools rebuild readable source from minified output. Appending a source-map comment and wrapping output in an immediately invoked function expression is standard webpack and esbuild behavior, and the rule title describes exactly that pattern. The file never loads at runtime, so it cannot execute anything in a user's browser.

The networking findings, including NET-XMLHTTPREQUEST-abp-background.js-14892, NET-FETCH-abp-background.js-18966, NET-FETCH-abp-background.js-56847 and their siblings, mark call sites inside the background worker and in UI helpers such as adblock-uiscripts-load_wizard_resources.js and button/help.js. A blocker needs fetch to pull subscription filter lists, check for updates and load its help panels. None of those call sites resolved to a suspicious host, and the endpoint extractor returned an empty list, so there is no monetization or exfiltration domain to point at here.

The two obfuscation flags, OBFUSCATION-LARGE_BASE64-abp-background.js-76950 and OBFUSCATION-FUNCTION_INDIRECT-abp-background.js-5471, sit in that same minified bundle. Large base64 literals in a blocker usually hold embedded icons, locale strings or rule fragments, and indirect function calls come from the dispatch tables minifiers generate. Neither flag pairs with a credential-domain indicator, a proxyware relay or a hijacked search template.

A skeptic would say the supply-chain rule fired at high severity, and one hit is enough to distrust a build. That rule matched a source map, which never executes, while the extension's manifest requests nothing and the endpoint list stays empty. Browser hijackers and proxyware in this pipeline carry search-domain templates, cookie access and relays to third-party hosts. We see none of those. The finding volume tracks one build artifact plus a handful of ordinary fetch calls.

What would change this verdict is a source map that resolves to injected code, a fetch destination outside AdBlock's own services, or a permission string the manifest does not need. Today the evidence shows a large, long-published blocker scanned at the artifact layer.

Key Reasons

• YARA--supply_chain_sourcemap_appended_iife matches /abp-background.js.map line 1, a build-time source map that never executes
• Network findings are fetch and XHR call sites in abp-background.js and helper scripts, with an empty endpoint list and no suspicious host
• OBFUSCATION-LARGE_BASE64 and OBFUSCATION-FUNCTION_INDIRECT both sit in a minified bundle, which is expected output
• Manifest declares empty permissions and empty host_permissions, giving no capability for credential, cookie or proxy abuse
• Publisher identity ([email protected], AdBlock, 65M users) is consistent with the long-running legitimate extension

False Positive Considerations

• Supply-chain YARA rule firing on an appended source map artifact
• fetch/XHR call sites counted as network findings with zero resolved suspicious domains
• large base64 and indirect function flags in minified bundle output
• empty permission and endpoint extraction from the manifest

Reviewed 2026-10-17; recommended action: suppress false positive; model confidence 92%.