OpenAI ChatGPT Computer History Feature on macOS Could Expose Detailed User Activity Logs to Infostealers
OpenAI has rolled out a new Computer History feature in the ChatGPT desktop application for macOS that converts recent user activity into structured summaries and memories for use by ChatGPT and Codex.
The system captures application switches, mouse clicks, text input, keyboard shortcuts, and additional context obtained through macOS accessibility APIs. These raw events are processed locally for up to 48 hours before deletion, after which brief summaries are generated and stored as ordinary Markdown files on the device.
The feature is turned off by default and requires users to enable Memories. It is currently available only to Pro, Business, and Enterprise subscribers, with administrator approval needed for corporate plans. Access is blocked in the European Economic Area, Switzerland, and the United Kingdom.
Users can choose which applications and websites participate, pause data collection at any time, or delete records for the last ten minutes, hour, day, or entirely. Screenshots, audio, and private browsing activity are explicitly excluded from the logs.
Although OpenAI states that raw events are not retained on its servers and are not used for model training, the generated Markdown memory files remain unencrypted. Any other application running under the same macOS user account can therefore read them, giving infostealers a ready-made daily activity report.
OpenAI additionally cautions about prompt injection risks: hidden instructions embedded on websites or inside applications may enter the context window and cause ChatGPT or Codex to follow attacker-controlled directives.
Related articles
AirTag and SmartTag: How to Detect If a Bluetooth Tracker Is Following You
Phones can now warn users when an unknown Bluetooth tracker such as an Apple AirTag or Samsung SmartTag is moving alongside them for an extended period. The alerts do not automatically confirm stalking, since the same tags can be left in taxis, rental cars or borrowed bags, yet they should never be ignored. AirTag and SmartTag rely on crowdsourced Bluetooth networks rather than built-in GPS or cellular connections, allowing them to report approximate locations only when nearby phones relay the signal. The article details differences from traditional GPS trackers, explains how to interpret notifications on iOS 17.5+ and Android 6.0+, and provides step-by-step guidance for locating hidden devices in clothing, vehicles or personal items. It also covers immediate safety actions, evidence preservation for police reports, and practical steps to reduce future tracking risks through account and permission hygiene.
VPN Encryption Alone Cannot Hide Traffic from Network Detection Systems
A Habr user known as mr_tom detailed why encrypted VPN connections remain detectable despite their encryption. Observers can identify connections through visible metadata such as server IP addresses, ports, transport protocols, handshake patterns, packet sizes, timing intervals, and overall flow behavior. Simple DPI systems block traffic by restricting known IPs or ports, while advanced filters build traffic fingerprints and use active probing to verify suspicious endpoints. The popular combination of VLESS, XHTTP, and REALITY operates across different layers rather than functioning as three equivalent VPN protocols. Even traffic on port 443 can be distinguished from standard HTTPS by analyzing handshake details and subsequent packet behavior. The core conclusion is that no universally unblockable VPN exists, as detection can rely on IP blocking, new signatures, or active verification regardless of encryption strength.
Browser Extension Anonymizes Sensitive Data Before Sending to AI Chatbots
A new browser extension automatically detects and replaces sensitive information such as names, INN numbers, bank cards, and emails with pseudonyms before any text or files reach AI chat services. The tool operates entirely locally in the browser, ensuring original data never leaves the user's device while allowing AI models to process anonymized placeholders. Responses from the AI are decrypted back to readable form only on the user's screen using an in-memory mapping that disappears when the browser closes. The solution supports over 70 file formats including DOCX, XLSX, and PDF, plus offline OCR for scanned documents to handle the most common leakage vectors used by HR, legal, and accounting teams. Multiple validation layers including checksums, Luhn algorithm, entropy checks, and normalization for Russian name declensions reduce false positives that plagued earlier regex-based attempts. The extension is available in a free version on the Chrome Web Store covering 33 data categories, with advanced file and scan features offered on request.
Browser Privacy Ranking 2026: Tor Browser Leads at 9.5 While Chrome and Yandex Rank Lowest
SecurityLab.ru has published a detailed 2026 browser privacy ranking that evaluates major browsers across six criteria including initial network behavior, site isolation, fingerprint resistance, funding model, configurability, and code auditability. Tor Browser scores highest at 9.5 for its comprehensive protections including unified fingerprinting and layered JavaScript controls, followed by Mullvad Browser at 9.0 and LibreWolf at 8.5. Brave earns 7.5 for built-in tracker blocking but faces criticism over default telemetry and cryptocurrency features. Firefox scores 6.0 out of the box yet reaches 8.5 after extensive configuration, while Google Chrome, Microsoft Edge, Opera, and Yandex Browser occupy the bottom positions due to persistent tracking mechanisms and closed-source components. The report also highlights the Local Mess localhost tracking technique used by Meta and Yandex that bypassed browser isolation entirely.