Docx Anonymizer: Open-Source Chrome Extension Uses Fine-Tuned rubert-tiny2 to Anonymize Personal Data Before AI Uploads
A new open-source Chrome extension called Docx Anonymizer allows users to automatically anonymize personal data in documents before uploading them to public AI services such as ChatGPT. The tool was developed to reduce the risk of violating Russia’s 152-FZ personal data protection law when employees send long contracts or other files containing names, organizations, and identifiers to cloud-based language models.
According to telemetry from the Solar Group, the volume of corporate data sent to public neural networks in Russia grew 30-fold in 2025, including legal documents. In one notable case, a Moscow court upheld the dismissal of a sales director who had uploaded internal reports to DeepSeek, resulting in the loss of a multi-million-ruble severance claim. These incidents highlight the need for lightweight, local anonymization solutions that do not themselves rely on cloud services.
The extension was built as a pure client-side JavaScript Chrome extension that intercepts text on any AI chat page, runs named-entity recognition locally, and replaces detected entities in real time. It never makes network requests, contains no telemetry, and can be installed without the Chrome Web Store. The developer deliberately avoided complex desktop applications to keep the workflow as short as possible.
The core model is a fine-tuned version of rubert-tiny2, a compact Russian BERT variant that runs in the browser through ONNX Runtime and WebAssembly. After int8 quantization the model occupies only 30 MB. Cold start takes approximately 0.3 seconds, and a 20-page contract is processed in about 1.7 seconds on a single thread inside a worker.
Training involved two iterations. The first used WikiANN data plus synthetic contract templates and knowledge distillation from a larger xlm-roberta-base teacher model. The second iteration added factRuEval news texts and 35 % synthetic business documents covering commercial proposals, invoices, meeting minutes, and chat messages. Negative examples such as capitalized headings and product names were explicitly included to reduce false positives on organizations. Final F1 scores on held-out data reached 0.913 for persons and 0.756 for organizations on WikiANN Russian test, with significantly improved results on real business genres.
The extension currently supports PDF files that contain a text layer; scanned documents are not processed. While organization names remain the most challenging class, the interface allows quick manual verification of any replacement. The full source code is available on GitHub under an open license, and the developer welcomes contributions of additional datasets or synthetic examples for further model improvement.
Related articles
AWG Router Delivers Script-Free AmneziaWG Cascades with Load Balancing, Speed Limits and Leak Prevention
AWG Router is an open-source Python-based web panel that manages complex AmneziaWG VPN cascades without hundreds of lines of bash scripts. The tool generates nftables rules, ip rules, and tc queues to route Russian networks directly while sending other traffic through foreign AmneziaWG servers or SOCKS5 proxies. It enforces strict leak protection so that traffic never falls back to the entry server’s Russian IP when an upstream tunnel fails. The panel supports per-client routing policies, domain-based rules via SNI inspection, automatic load balancing across multiple exits, and granular speed limits using HTB and fq_codel. All state is stored in resilient JSON files, and the system continues operating even if the panel itself is restarted or updated. The project is released under the MIT license with full source available.
TV Manufacturers Admit They Cannot Fully Disable User Data Collection
The Verge surveyed 11 major television and streaming device makers about their data collection practices, receiving responses from seven companies including Amazon, Apple, Google, LG, Roku, Samsung and TCL. Automatic Content Recognition (ACR) technology, used for targeted advertising and audience measurement, remains a central concern, with Samsung, LG and Amazon disabling it by default. Google enables diagnostic and advertising data collection under standard settings, while Apple stated that Apple TV 4K does not employ ACR. Amazon, Google, Samsung and Roku acknowledged that complete opt-out from data transmission is impossible, with devices caching information offline for later upload. Research from Northeastern University revealed that Vizio continued sending data even after users declined collection, labeling it as operational telemetry. LG was found to potentially store unencrypted voice command transcripts and determine location via nearby Wi-Fi networks when ACR is active.
Personal Data in Russian Franchise Networks: Who Qualifies as Data Operator Under 152-FZ
In typical franchise models, customer personal data is collected at franchisee points, stored in the franchisor’s central system, and used for network-wide loyalty programs and marketing. The key legal question is which party determines the purposes of processing and therefore bears operator status under Federal Law No. 152-FZ. The law defines an operator as the entity that independently or jointly sets processing purposes, data composition, and actions, regardless of contractual labels. Franchisees usually act as operators when concluding contracts with customers, while franchisors become processors unless they pursue independent goals such as running loyalty schemes or sending branded advertising. When a franchisor defines its own purposes, both parties may simultaneously qualify as operators for different processing objectives, requiring separate legal bases and documentation. Proper data-processing instructions must be embedded in the commercial concession agreement and must list specific data categories, actions, security measures, and breach-notification timelines to avoid administrative fines ranging from 150,000 to 500,000 RUB.
Dynamic Sites on Yandex Cloud Storage Enable Communication Under White-List Restrictions
A developer facing mobile internet white lists explored ways to exchange encrypted messages without relying on blocked infrastructure. Initial attempts with a national messenger bot failed due to poor API documentation and account hijacking. The team then modeled the problem as three neighboring plots where Alice and Bob route messages through Eve while a slow postman handles delivery. Yandex Disk was tested as a message store accessible via direct GET and POST requests, though it required accounts and suffered from 15-25 second latency. The final approach used Yandex Cloud object storage, whose domains remained whitelisted, allowing HTML files to load and function even when only government services and select platforms were reachable.