MERLIN BOT

AI Security Review

Evidence context: threat category none; evidence quality moderate.

MERLIN BOT is a CRM that sits on WhatsApp Web, published under the address [email protected]. The scan covers version 7.4.3.87 and returns 1,195 findings. Almost all of them are the kind of noise a bundled content script produces.

Start with what the code does. content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28912.js and the file ending 175.js each open a Socket.IO client. That is how a live inbox receives messages without polling, and it fits a product whose stated job is organizing WhatsApp chats. background.js:137 makes a fetch call, as do the twelve other NET-FETCH hits spread across the v_7_4_3 chunk family. Two files trigger NET-JQUERY_AJAX because they bundle jQuery. Each of these findings marks a transport mechanism inside a bundled library rather than a destination.

The endpoint list is where the scan gets messiest. It contains border.radius.circle, color.border.accent.orange, btn.glass, calls.function.arguments.delta, artboard.phone and above.sm, which are CSS token names and JavaScript property chains misread as hostnames. api.statsigcdn.com is a feature-flag CDN used by many front ends. app.wascript.com is the one entry that reads like a real backend belonging to this product. The 415 indicators of compromise are that same extractor output at volume, and the 755 code-smell items are rules that fire on any non-trivial minified JavaScript. Neither group carries signal here.

Two obfuscation findings need a direct answer. OBFUSCATION-FUNCTION_INDIRECT fires in content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d289260.js, and OBFUSCATION-LARGE_BASE64 fires in funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js. Both files carry content-hash names, the pattern webpack and esbuild emit. Indirect calls and embedded base64 blocks are ordinary build output. Across the whole package, zero malware signatures matched, and that check is what separates a packed loader from a packed build.

The strongest counterargument is that "bundled" is an assumption about build tooling, and a large base64 block can hold anything. That objection has force. Two specific things would close it: the decoded content of funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js, and the registrant behind app.wascript.com. Until those are checked, intent stays unproven in both directions. What the evidence does show is a small extension with 17 users whose network findings all land on library boilerplate, with no signature and no domain aimed at a login page or a financial site.

One limitation: the scan records no permission strings for this extension, so API access cannot be judged from the manifest in this data. What any WhatsApp CRM can see is your chat list and message text, which comes with the product category rather than from a hidden capability.

Key Reasons

• Zero malware-signature and zero malware findings across 1,195 results in version 7.4.3.87.
• Network findings are library sinks: Socket.IO in content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28912.js, jQuery Ajax, and fetch at background.js:137.
• The IoC set is extractor noise, with CSS tokens and property chains such as border.radius.circle and calls.function.arguments.delta misread as hostnames.
• Both obfuscation hits sit in content-hash filenames consistent with webpack or esbuild output rather than hand-written packing.
• 755 code-smell items are low-severity rules that fire on any minified JavaScript.

False Positive Considerations

• IoC extractor misreading CSS tokens and JavaScript property chains as domain names
• 755 code-smell rules matching minified and bundled JavaScript
• Minifier output (indirect calls, embedded base64) reported as obfuscation
• Per-file duplication of library network sinks (Socket.IO, jQuery Ajax, fetch) across bundled chunks

Reviewed 2026-10-14; recommended action: no action; model confidence 75%.