Build1 distinct publisher3 min readUpdated
Two obfuscated Alibaba anti-bot scripts on AliExpress built silent Web Audio graphs with no media element. Tab, browser and OS mute did nothing. Closing the tab worked.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
A technical analysis published August 20th reports that AliExpress loaded two obfuscated Alibaba security scripts which built silent Web Audio processing graphs, and that on one user's Windows setup those graphs stopped multipoint Bluetooth headphones from switching audio back to a phone [1]. Because the scripts generated and analyzed audio without an `<audio>` or `<video>` element, muting the AliExpress tab, muting Firefox and muting Windows all failed to release the computer's audio connection; closing the tab released it immediately [2].
The route to the finding is worth the detail. According to the analysis, published by the operator of the laserphile blog, identified on the site as m-c-tech, the investigation started with a symptom rather than a suspicion: a pair of multipoint headphones repeatedly stopped playing phone audio while an AliExpress page was open on a connected PC, several seconds after page load, and recovered the moment the tab closed [3][5]. An initial inspection turned up no media elements, no playback calls, no active Media Session and no visible content that would explain it [6]. The author then wrapped the browser's `AudioContext` constructor and `AudioNode.connect()` to log audio activity [7]. The homepage was running two audio contexts, and stack traces pointed at Alibaba-hosted files named collina.js and fireyejs.js, both under an AWSC directory associated with Alibaba's browser security tooling [8].
The graph itself is the point. Per the analysis, both scripts built the same shape: a sawtooth oscillator into an analyzer and a script processor, then a gain node set to zero, then a connection to the system audio destination [11]. Zero gain makes the output inaudible, but the destination connection still makes the browser push the graph through the machine's audio path, and on this setup that was enough to hold the PC side of the multipoint link open [12]. The Web Audio specification treats the destination as the final audio output, and browser documentation notes it commonly maps to speakers or another physical output device, while a gain of zero effectively mutes the signal [13]. So the user had three mute controls, all operating on volume, against a process that was already silent; the only lever that worked was killing the page [17].
Autoplay policy does not catch this either. Browser autoplay protections cover Web Audio, but their behavior depends on settings, prior interaction and whether the audio counts as inaudible, and MDN's current guidance says muted or zero-volume media may be permitted automatically [14].
Alibaba Cloud's documentation describes AWSC JavaScript components as part of its no-interaction verification product, and its anti-bot documentation says browser-side collectors gather environmental characteristics, automation indicators and user behavior to separate humans from bots, injected across every page of a protected site [9]. An Alibaba developer article identifies fireye.js as part of an AWSC security system for anti-scraping, anti-abuse and human-versus-bot detection, and says calls within the script collect low-level hardware information and can consume significant processing resources [10]. The wider script inspection found code reading canvas output, WebGL properties, screen dimensions, hardware concurrency, device memory, supported media formats, WebRTC behavior, timing information and user interaction, plus code to encrypt and transmit results to Alibaba telemetry services [15]. Those measurements match established browser-fingerprinting techniques [16].
The honest caveat: this documents one configuration, not a test across headphones, operating systems and browsers [3].
What to watch: whether anyone reproduces the handoff failure on other headphone, OS and browser combinations; whether browser vendors stop treating a destination-connected zero-gain graph as an active audio session for the operating system's purposes; and whether the AWSC scripts keep the destination connection, which the analysis suggests is not needed to compute a fingerprint from the analyzer and script processor [11].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
According to the laserphile analysis, both scripts constructed a similar processing chain: a sawtooth oscillator fed an analyzer and script processor, followed by a gain node set to zero and a connection to the system audio destination.
AliExpress loaded two obfuscated Alibaba security scripts that created silent Web Audio processing graphs and, on one user's Windows setup, prevented multipoint Bluetooth headphones from switching audio back to a phone, according to a technical analysis published August 20th.
The zero-gain setting made the result inaudible, but the final destination connection still caused the browser to process the graph through the computer's audio path; on the author's setup that activity was enough to keep the PC side of the multipoint connection active, blocking the headphones from returning to phone audio.
The scripts generated and analyzed audio without an <audio> or <video> element, so muting the AliExpress tab, Firefox or Windows did not release the computer's audio connection. Closing the tab did.
The Web Audio specification defines an audio context as a graph of connected processing nodes, with the destination representing the final audio output; browser documentation notes that a destination commonly maps to speakers or another physical output device, while a gain value of zero effectively mutes the signal.
The investigation began after a pair of multipoint headphones repeatedly stopped playing audio from a phone while an AliExpress page was open on a connected PC; the interruption appeared several seconds after the page loaded and ended immediately when the tab closed.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Traceable method, single configuration
The mechanism is documented with named scripts, an instrumentation method another engineer could repeat, and corroboration from Alibaba Cloud's own product documentation, the Web Audio specification, MDN and W3C security guidance. Against that, the hardware outcome rests on one Windows/Firefox/multipoint-headset configuration reported by a single pseudonymous blogger, relayed by a single publisher, with no vendor response and no cross-platform reproduction.
Code widely deployed, side effect unquantified
Adoption of the underlying technology is real and consequential: the anti-bot collectors run in production on AliExpress's homepage, and Alibaba Cloud markets site-wide injection of the same components for protected sites. What is not measured is the reach of the harmful behavior -- exactly one user's device configuration is shown to lose Bluetooth handoff, with no data on other browsers, headsets or user counts.
Slight over-reach, disclosed
Framing runs modestly ahead of proof: an n=1 hardware symptom is presented as an infrastructure-level flaw in anti-bot design. The overstatement is small because the headline and body both scope it to one user, the article states what the browser-side evidence does not establish about Alibaba's use of the data, and the mechanism -- zero-gain graph still connected to the system destination -- is documented rather than asserted.
Mixed independent and vendor self-description
The primary analysis comes from an independent, pseudonymous blog with no commercial stake disclosed in the material, which limits obvious distortion, though pseudonymity also removes accountability. The corroborating material is vendor self-description -- Alibaba Cloud product documentation, an Alibaba developer article and AliExpress's privacy policy -- which is written to promote or justify the tooling, and no party with a direct interest was asked to respond.
Mechanism solid, generality unproven
High confidence that the scripts exist, build zero-gain graphs connected to the audio destination, and collect and transmit fingerprint-like signals -- that is documented with named artifacts and vendor corroboration. Lower confidence that this reliably breaks Bluetooth multipoint handoff in general, since the outcome rests on one configuration, one blogger and one publisher with no vendor or browser-maker input.
product
Alibaba's 75% profit drop is what a working cloud business costs1 distinct publisher
build
Anti-bot systems now score the session, which means your proxy pool is not a mitigation1 distinct publisher
build
"Follow the same standard" is how mobile accessibility debt gets onto the books1 distinct publisher
build
The fourth Web Push requirement: iOS will not deliver until the user installs your site1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 20, 2026