Yandex Details Alice Voice Assistant Audio Buffering and Data Handling in Android Apps
Yandex has published a comprehensive technical reply to recent scrutiny of its Android applications, addressing allegations of excessive data collection including audio, contacts, and banking information.
The company confirmed that the Alice voice assistant maintains a cyclic audio buffer of roughly 1.5 seconds prior to the activation phrase and 0.5 seconds afterward. This buffer operates in memory when the app is open, and a portion of the audio preceding a command may be sent to servers to evaluate speech recognition quality. Yandex noted that remote adjustment of buffer size could raise user concerns and pledged to impose strict limits in an upcoming release.
Regarding contacts, Yandex stated that access supports voice commands such as “Call mom.” On initial synchronization the entire address book may be transmitted to servers, after which only modifications are sent. Names and numbers are not hashed because raw values are required for accurate speech recognition.
The company emphasized that its applications do not read message content from WhatsApp, Telegram, or Viber. They only observe contact details that these messengers publish to the system address book on Android.
Accusations concerning insecure handling of bank cards were dismissed: card details are routed directly to an isolated PCI DSS environment rather than ordinary application servers. The VPN interface was attributed to network diagnostics, the backup DNS to bypassing provider outages, and the list of other applications in the manifest to a standard Android mechanism.
Related articles
One Request, Five Observers: What Websites, Providers, DNS and VPNs Learn When Loading a Page
The article breaks down exactly what each participant in a typical web request can observe when a user visits a page over HTTPS. It examines the roles of the browser, DNS resolver, ISP, VPN service and the destination site itself, showing that each sees different pieces of metadata or content. HTTPS protects the page body and parameters from network observers, while Encrypted Client Hello and secure DNS further limit visibility of domain names. VPNs replace the user's home IP address with the VPN exit node but introduce a new trusted party that sees all traffic metadata. Browser fingerprinting, cookies and account logins often allow sites to re-identify users even after an IP change. The piece stresses that privacy tools must be chosen according to the specific threat model rather than relying on any single mechanism.
Windscribe Releases Open-Source PowerShell Script to Remove Microsoft's Persistent Global Device Identifier from Windows
Windscribe has published an open-source PowerShell script called deGDID that removes Microsoft's Global Device Identifier (GDID) from Windows systems and prevents the creation of new identifiers. The persistent tracker survives IP address changes and operates below the VPN layer, allowing Microsoft to maintain device tracking even when users employ privacy tools. The script was developed after the FBI used GDID to identify a suspected hacker, raising concerns about undisclosed device fingerprinting. deGDID modifies registry access control lists and blocks the DeviceAdd interface to stop Windows from regenerating the identifiers after reboots or server contacts. While effective at disabling the tracking mechanism, the tool can break authentication to login.live.com and disrupt certain Microsoft cloud services. It is intended only for unmanaged consumer devices and will not run on domain-joined or corporate-managed systems. The project is described as experimental and cannot delete identifiers already transmitted to Microsoft servers.
Yandex Apps Leak Pre-Trigger Audio Buffers, Payment Data, and Contacts via Reverse Engineering
Reverse engineering of Yandex Search and Yandex Browser APKs reveals extensive data collection practices on Android devices. The apps maintain a server-controlled pre-trigger audio buffer that captures up to three seconds or more of microphone input before the wake word Alice is detected. WiFi fingerprinting, full contact book synchronization via ContentObserver, and transmission of PAN and CVV details to mobpayment.yandex.net occur before tokenization. Additional findings include 94 JavaScript Bridge methods, logcat exfiltration with AES encryption, hardcoded Yandex DNS servers, and a native surveillance library named libquarkenstein_daemons.so. The analysis also covers passive geolocation, cell tower data collection, and inventory of installed applications including competitors such as Chrome, Firefox, WhatsApp, and Telegram. These mechanisms operate under remote configuration flags and bypass several Android privacy restrictions through manifest queries.
TSPU Filtering Disrupts Timeweb Cloud Servers: Diagnosis, CDN Failures, and Reverse Proxy Bypass
Russian hosting provider Timeweb Cloud experienced widespread TSPU-based DPI filtering starting in early June that selectively blocked TLS handshakes on port 443 while leaving SSH, ICMP, and TCP connectivity intact. The issue affected multiple providers including Beget and Selectel, was publicly acknowledged by Timeweb on June 5, and proved highly variable by region, operator, and time of day. Attempts to mitigate via new IP addresses or Timeweb's own CDN failed due to poisoned caches and platform outages, while Yandex Cloud CDN blocked all POST requests required for WordPress functionality. A working solution involved deploying a minimal nginx reverse proxy on another Russian cloud VPS that preserves full HTTP methods, handles certificate synchronization, and routes ACME challenges correctly. Timeweb support later closed tickets without resolution after requesting ineffective mtr traces that cannot detect DPI behavior. The case highlights systemic challenges in diagnosing state-mandated filtering and the limitations of standard network diagnostics against selective TLS interference.