InfiChat

AI Security Review

Evidence context: threat category none; evidence quality moderate.

InfiChat's store listing describes a CRM layered over WhatsApp Web: chat organisation and sales tooling inside the browser tab. The package ships as hashed build artefacts under content/assets/js/ and funnil/assets/js/, the layout a webpack or Vite build leaves behind. The manifest declares no permission strings and no host permissions, so nothing in the package asks the browser for cookies, downloads, the clipboard or history.

The network detections are library fingerprints. NET-JQUERY_AJAX at content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28923.js:1 fires because jQuery's request helper is bundled; NET-SOCKET_IO at content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d28912.js:15 fires because the socket.io client is bundled. A shared inbox that keeps conversation state in sync across agents needs a websocket, and socket.io is the common way to get one. NET-FETCH at background.js:137 is the same mechanism sitting in the service worker.

The 430 indicator matches are mostly noise. The endpoint list mixes real infrastructure (api.statsigcdn.com, Statsig's feature-flag CDN, plus the Atlassian API hosts that arrive with a bundled dependency) with CSS design tokens such as border.radius.circle, btn.glass and color.border.accent.orange, and with API constants such as calls.function.arguments.delta that exist as property names inside a bundled SDK. Short entries like above.sm, ae.jobs and ae.online come from strings split around two-letter country suffixes. None of these resolve to a server that receives chat data.

Two obfuscation matches carry more weight on their face. OBFUSCATION-LARGE_BASE64 points at funnil/assets/js/9cf42769-2ad5-4714-8614-9a2b4c455349.CgnacD38.js, an inline base64 payload, and OBFUSCATION-FUNCTION_INDIRECT points at content/assets/js/v_7_4_3_87_0a650f0a-57a6-4989-860e-3a9d8748d289260.js. Build tools embed base64 for icons, fonts and source maps, and function indirection is what a minifier produces when it rewrites a direct call. Both files sit in directories alongside dozens of sibling bundles from the same build, which is what compiled output looks like.

The strongest argument against this reading is that a WhatsApp CRM reads the messages in the tab. A content script on web.whatsapp.com sees every conversation rendered, so the socket.io channel or the fetch in background.js could ship chat content to a server outside the vendor. The empty permission fields do not rule that out on their own, because content_scripts lives under its own manifest key and the bundle does not expose the socket.io host it connects to. Against that, the package contains zero malware matches and zero malware-signature matches, no credential-harvesting findings, and no indicator aimed at a login domain such as a bank or mailbox. The only named third-party hosts are Statsig and Atlassian, both ordinary in bundled SDK and feature-flag code. A build with harmful intent usually leaves at least one of those traces, and this one leaves none.

The finding volume reflects bundled JavaScript and a noisy extractor, not behaviour aimed at the person using it.

Key Reasons

• Zero malware and zero malware-signature findings across 1209 matches; the volume sits in ioc (430) and code-smell (754)
• NET-JQUERY_AJAX, NET-SOCKET_IO and NET-FETCH map to bundled jQuery, the socket.io client and background.js:137, all expected in a live-chat CRM
• Reported endpoints are CSS design tokens (border.radius.circle, btn.glass), SDK property names (calls.function.arguments.delta) and ordinary infrastructure (api.statsigcdn.com)
• Obfuscation matches are an inline base64 asset and a minifier indirect-call pattern inside hashed build output under content/assets/js/ and funnil/assets/js/
• Manifest lists no permissions and no host permissions

False Positive Considerations

• XIOC extractor read CSS design tokens and SDK property names as domains (border.radius.circle, color.border.accent.orange, calls.function.arguments.delta)
• Code-smell rules fired on bundled JavaScript, producing 754 low-severity matches with no malware signatures behind them
• Network findings are library fingerprints (jQuery Ajax, socket.io, fetch) inside hashed build artefacts
• Obfuscation findings match standard bundler output: an embedded base64 asset and minified function indirection

Reviewed 2026-10-11; recommended action: suppress false positive; model confidence 75%.