DigitaZap

AI Security Review

Evidence context: threat category none; evidence quality moderate.

DigitaZap's manifest declares no permissions and no host permissions. That matters, because permissions set the outer limit of what an extension can touch, and this one asks for nothing beyond running its own content scripts inside the page. Whatever it does to WhatsApp Web, it does with ordinary web APIs from a content script.

The network findings are library-level detections inside bundled files. background.js:137 makes a fetch call. content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28912.js, ...49149.js, ...49175.js and ...4966.js each open a socket.io connection, at lines 15, 64, 37 and 1. The remainder are NET-JQUERY_AJAX and NET-FETCH hits scattered across similarly named bundles under content/assets/js/. A CRM that keeps a chat list in sync needs a socket and a REST call, and that is what these lines are. Detections of this shape, one per minified vendor file, are what bundled third-party code looks like.

The endpoint list needs reading rather than counting. app.wascript.com is the single entry that resembles a product backend, and a service domain for a WhatsApp tool is exactly where you would expect a CRM of this kind to send its data. The other nineteen entries are extraction artifacts. above.sm, below.sm, ae.jobs, chat.live, artboard.phone and avatar.online were lifted from minified string concatenation. border.radius.circle, color.border.accent.orange and calls.function.arguments.delta are CSS token names and JavaScript property paths that no resolver would ever treat as hosts. Seeing property chains in the endpoint list is a reliable sign the extractor was scraping strings, not following traffic.

Two obfuscation findings carry medium severity: OBFUSCATION-LARGE_BASE64 on funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js, and OBFUSCATION-FUNCTION_INDIRECT on content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d289260.js. Neither sits alongside a malware signature, which the triage rules require before a base64 blob in build output is treated as a payload. Bundlers inline images, fonts and config objects constantly, and that produces base64 of this size.

Zero malware signature findings in 1195 total. No credential-harvesting code, no proxyware, no browser hijack. The 754 code-smell matches come from the same bundled JavaScript that every non-trivial web app ships.

A skeptic has a fair case. The publisher is listed as [email protected], an email on a generic store domain, with no company name. The extension has 14 users. A CRM for WhatsApp reads private conversations by design, and a large base64 blob is what a hidden payload would also look like. What tips the balance is the absence of any concrete capability: no permissions, no malware signature anywhere in the tree, and network calls that resolve to fetch, socket.io and jQuery AJAX inside vendor bundles. The right follow-up is a re-scan after the next version bump, when any change of behaviour would show up.

Key Reasons

• Zero malware-signature and zero malware findings across 1195 total; all severity is medium or low.
• All 22 network findings are NET-FETCH, NET-SOCKET_IO or NET-JQUERY_AJAX inside bundled files under content/assets/js/ and background.js, matching real-time chat sync behavior.
• Manifest declares no permissions and no host permissions, so the extension has no privileged capability.
• Endpoint list is dominated by extractor artifacts such as border.radius.circle, color.border.accent.orange and calls.function.arguments.delta, which are CSS tokens and property paths.
• Two medium obfuscation hits (OBFUSCATION-LARGE_BASE64, OBFUSCATION-FUNCTION_INDIRECT) appear without any co-located malware signature.

False Positive Considerations

• Minified vendor bundles under content/assets/js/ producing one network detection per file
• IoC extractor scraping minified strings and reporting CSS tokens and property paths as domains
• Code-smell rules firing on bundled third-party JavaScript (754 low-severity matches)

Reviewed 2026-10-13; recommended action: monitor; model confidence 72%.