ZAPGYN

AI Security Review

Evidence context: threat category none; evidence quality moderate.

ZAPGYN adds CRM features to WhatsApp Web: chat organization, contact handling and campaign tools. The bundle is a large JavaScript application, and almost everything flagged inside it traces to bundled libraries or to the extractor misreading UI strings as network destinations.

Start with what did not match. Across the scan there are no malware-signature hits, no malware-family hits, no exposed secrets and no tool-poisoning findings. On a bundle this size that absence carries more weight than the 754 code-smell entries, which fire on ordinary JavaScript constructs.

The 22 network findings are library detections, not evidence of exfiltration. NET-JQUERY_AJAX-content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb223.js, NET-SOCKET_IO-content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2148.js:64 and NET-FETCH-background.js:137 mark where jQuery's ajax helper, the socket.io client and the fetch API appear in the source. Those are the plumbing inside any bundled web app. Alone, they say nothing about where data goes or who controls the far end.

The endpoint list is the clearest sign of extractor noise. Entries like ae.jobs, above.sm, below.sm, btn.glass, avatar.online, border.radius.circle, artboard.phone and color.border.accent.orange are not hostnames a server could resolve. They are UI tokens, class-name fragments and JSON keys lifted out of the source. Mixed in are api.statsigcdn.com for feature-flag delivery and api.dev.atlassian.com plus atlassian-isolated.net for editor and collaboration libraries. One endpoint, app.wascript.com, belongs to a WhatsApp automation backend, which is what a WhatsApp CRM talks to.

The two obfuscation findings are routine for this kind of build. OBFUSCATION-LARGE_BASE64-funnil/assets/js/ed6113f2-bc09-4421-a453-bbbd66fe9488.DCssQ7AT.js points at a content-hashed chunk with inlined assets, and OBFUSCATION-FUNCTION_INDIRECT-content/assets/js/v_7_4_3_86_085ff17f-7684-4a41-bfdd-8bd9d5293bb2259.js is the indirect-call pattern webpack and Babel emit. Neither is a hand-written packer. The recorded manifest lists no permission strings and no host permissions, so reach comes from content scripts injected into WhatsApp Web rather than broad browser access.

The strongest argument against calling this clean is that a WhatsApp CRM runs inside your chat window. Its content scripts can read message text, and the socket.io and fetch calls in content/assets/js/ can move that data to a server while you work. That capability is real and it is the product. What is missing is any evidence the data path goes somewhere unexpected: no IoCs against login or banking domains, no cookie-access pattern, no obfuscated entry point, and the only non-CDN endpoint is the extension's own backend. A runtime capture of traffic to app.wascript.com would close that question.

Key Reasons

• Zero malware-signature, malware, secret and tool-poisoning matches across 1183 findings; only medium and low severity entries exist.
• All 22 network findings are library-level detections (NET-JQUERY_AJAX, NET-SOCKET_IO, NET-FETCH) inside content/assets/js and background.js, with no destination attribution.
• Endpoint list is dominated by non-resolvable UI tokens (btn.glass, border.radius.circle, artboard.phone, above.sm), a signature of XIOC extractor noise.
• Obfuscation findings are OBFUSCATION-LARGE_BASE64 and OBFUSCATION-FUNCTION_INDIRECT in webpack-style hashed bundle chunks, not a hand-written packer.
• The only non-CDN endpoint, app.wascript.com, is a WhatsApp automation backend consistent with the extension's stated CRM purpose.

False Positive Considerations

• 404 IoC entries include UI tokens, CSS class fragments and JSON keys misread as domains (ae.jobs, btn.glass, border.radius.circle, calls.function.arguments.delta).
• 754 code-smell findings generated by bundled third-party libraries in content/assets/js and funnil/assets/js.
• Obfuscation rules firing on webpack/babel output: large inlined base64 assets and indirect function-call patterns in content-hashed chunks.
• High finding volume from dist-style bundled dependencies rather than from any single flagged behaviour.

Reviewed 2026-10-10; recommended action: suppress false positive; model confidence 76%.