惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

T
Threat Research - Cisco Blogs
V
V2EX
爱范儿
爱范儿
Martin Fowler
Martin Fowler
T
The Blog of Author Tim Ferriss
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
C
Check Point Blog
罗磊的独立博客
A
About on SuperTechFans
MyScale Blog
MyScale Blog
S
Security @ Cisco Blogs
博客园 - 聂微东
Simon Willison's Weblog
Simon Willison's Weblog
Cyberwarzone
Cyberwarzone
云风的 BLOG
云风的 BLOG
U
Unit 42
Latest news
Latest news
Apple Machine Learning Research
Apple Machine Learning Research
Security Latest
Security Latest
Y
Y Combinator Blog
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
The Hacker News
The Hacker News
C
Cyber Attacks, Cyber Crime and Cyber Security
Cisco Talos Blog
Cisco Talos Blog
AWS News Blog
AWS News Blog
NISL@THU
NISL@THU
美团技术团队
S
Securelist
P
Privacy International News Feed
T
Tenable Blog
酷 壳 – CoolShell
酷 壳 – CoolShell
人人都是产品经理
人人都是产品经理
S
SegmentFault 最新的问题
Hugging Face - Blog
Hugging Face - Blog
cs.CV updates on arXiv.org
cs.CV updates on arXiv.org
K
Kaspersky official blog
Google Online Security Blog
Google Online Security Blog
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
www.infosecurity-magazine.com
www.infosecurity-magazine.com
D
Darknet – Hacking Tools, Hacker News & Cyber Security
B
Blog RSS Feed
W
WeLiveSecurity
A
Arctic Wolf
Hacker News - Newest:
Hacker News - Newest: "LLM"
Google DeepMind News
Google DeepMind News
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Recent Commits to openclaw:main
Recent Commits to openclaw:main
N
Netflix TechBlog - Medium
有赞技术团队
有赞技术团队
V2EX - 技术
V2EX - 技术

The Register - Security: Research

www.theregister.com Self-destructing Mistic backdoor linked to access broker selling corporate footholds to ransomware gangs PRC-linked spies hid inside medical and military networks for more than a year, snooping through Gmail and stealing data Nobody needs Mythos or 0-days to build a chaos-causing computer worm – free open source models work just fine ChatGPT blindly trusts browser content, turning the page into a payload Russia-linked threat group put ChatGPT to work from lure to payload Kids can bypass some age checks with a drawn-on mustache What type of 'C2 on a sleep cycle' do they leave behind? Novel Chinese spy group found in critical networks in Poland, Asia ORNL builds more sensitive GPS interference detector Researchers find sabotage malware that may predate Stuxnet Vibe coding upstart Lovable denies data leak, cites 'intentional behavior,' then throws HackerOne under the bus Anthropic, Google, Microsoft paid AI bug bounties – quietly Security reserchers tricked Apple Intelligence into cursing Don't open that WhatsApp message, Microsoft warns Security boffins harvest bumper crop of API keys from web Lightning-fast exploits mean patch fast, says Cisco Talos AI agents are 'gullible' and easy to turn into your minions Smooth criminals talking their way into cloud environments, Google says Snoops plant info-stealing malware on iPhones, Google warns Cybercrime up 245% since the start of the Iran war Rogue AI agents can work together to hack systems Fake applicants are sending security-killing malware AI agent hacked McKinsey chatbot for read-write access Kaspersky: No signs Coruna iPhone exploit kit made by US Perplexity Comet browser hole was exploitable via cal invite DEF CON hackers 'fed up with government,' Jake Braun says DEF CON hackers 'fed up with government,' Jake Braun says Ransomware payments cratered in 2025 – attacks did not Ransomware payments cratered in 2025 – attacks did not Claude's collaboration tools allowed remote code execution AI takes a swing at online anonymity Fake 'interview' repos lure Next.js devs into running secret-stealing malware Threat intelligence supply chain is full of weak links AI agents abound, unbound by rules or safety disclosures RAT disguised as an RMM costs crims $300 a month Android malware taps Gemini to navigate infected devices Posting AI caricatures on social media is bad for security Payroll pirates conned the help desk, stole employee’s pay Microsoft boffins show LLM safety can be trained away For the price of Netflix, crooks can rent AI crime ops For the price of Netflix, crooks can rent AI crime ops Fast Pair, loose security: Bluetooth accessories open to silent hijack Fast Pair flaw exposes Bluetooth devices to hijacking A simple CodeBuild flaw put every AWS environment at risk A simple CodeBuild flaw put every AWS environment at risk DeadLock ransomware uses smart contracts to evade defenders Python libraries in AI/ML models can be poisoned w metadata OpenAI patches déjà vu prompt injection vuln in ChatGPT Fake Windows BSODs check in at Europe's hotels to con staff into running malware Hotel staff tricked into installing malware by bogus BSODs Your car’s web browser may be on the road to cyber ruin China's Ink Dragon hides out in European government networks NCSC finds cyber deception tools work, if deployed right 10K Docker images spray live cloud creds across the internet 'Botnets in physical form' are top humanoid robot risk 'Botnets in physical form' are top humanoid robot risk Apache warns of 10.0-rated flaw in Tika metadata toolkit Novel clickjacking attack relies on CSS and SVG 'Exploitation is imminent' of max-severity React bug Swiss government bans SaaS and cloud for sensitive info Scattered Lapsus$ Hunters stress testing Zendesk weak spots HashJack attack shows AI browsers can be fooled with '#' New ClickFix attacks use fake Windows Updates to swipe creds Years-old bugs in open source took out major clouds at risk LLM-generated malware improving, but not operational (yet) 3.5B WhatsApp users' info scooped through enumeration flaw 3.5B WhatsApp users' info scooped through enumeration flaw 50k more ASUS routers pwned by evolving Beijing-linked op Overconfidence is the new zero-day as teams stumble through cyber simulations LLM side-channel attack could allow snoops to guess topic Landfall spyware used in 0-day attacks on Samsung phones MIT Sloan shelves paper about AI-driven ransomware Security hole slams Chromium browsers - no fix yet OpenAI Atlas Browser tripped up by malformed URLs Devs of VS Code extensions are leaking secrets en masse Chatbots that butter you up make you worse at conflict Tile trackers leak unencrypted Bluetooth data, say boffins Beijing's RedNovember hacked critical US, global orgs Lazarus RAT code resurfaces in North Korean IT-worker scams Suspected Chinese spies broke into 'numerous' enterprises Deepfaked calls hit 44% of businesses in last year: Gartner Kaspersky: RevengeHotels returns with AI-coded malware Ruh-roh. DDR5 memory vulnerable to new Rowhammer attack HybridPetya ransomware dodges UEFI Secure Boot
Browser 'privacy' extensions have eye on your AI, log all your chats
2025-12-17 · via The Register - Security: Research

Ad blockers and VPNs are supposed to protect your privacy, but four popular browser extensions have been doing just the opposite. According to research from Koi Security, these pernicious plug-ins have been harvesting the text of chatbot conversations from more than 8 million people and sending them back to the developers.

The four seemingly helpful extensions are Urban VPN Proxy, 1ClickVPN Proxy, Urban Browser Guard, and Urban Ad Blocker. They're distributed via the Chrome Web Store and Microsoft Edge Add-ons, but include code designed to capture and transmit browser-based interactions with popular AI tools.

"Urban VPN Proxy targets conversations across ten AI platforms," said Idan Dardikman, co-founder and CTO of Koi, in a blog post published Monday. 

The research firm said that the platforms targeted include ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, DeepSeek, Grok, and Meta AI.

"For each platform, the extension includes a dedicated 'executor' script designed to intercept and capture conversations," said Dardikman, who explained data harvesting is enabled by default through a hardcoded configuration flag. "There is no user-facing toggle to disable this. The only way to stop the data collection is to uninstall the extension entirely."

According to Dardikman, the Urban VPN Proxy extension monitors the user's browser tabs and, when the user visits one of the targeted platforms (e.g., chatgpt.com), it injects the "executor" script into the page.

"Once injected, the script overrides fetch() and XMLHttpRequest – the fundamental browser APIs that handle all network requests," he explained. "This is an aggressive technique. The script wraps the original functions so that every network request and response on that page passes through the extension's code first."

The script parses the intercepted API responses and then packages and transmits the data via window.postMessage to the extension's content script, along with the identifier PANELOS_MESSAGE. The content script then passes the data to a background service worker for exfiltration over the network to endpoints at analytics.urban-vpn.com and stats.urban-vpn.com.

The Register reached out to Urban VPN, affiliated company BiScience, and 1ClickVPN at their respective privacy email addresses. All three requests bounced.

Pointing to prior investigative material published by security researcher Wladimir Palant and John Tuckner of Secure Annex that details BiScience's collection of clickstream/browsing history data, Dardikman said his company's findings show BiScience expanding into the collection of AI conversations.

He notes that while Urban VPN does disclose AI data collection during the setup prompt and in its privacy policy, the Chrome Web Store listing indicates that data is not being sold to third parties outside approved use cases and that AI conversations are not specifically mentioned.

"The consent prompt frames AI monitoring as protective," he said. "The privacy policy reveals the data is sold for marketing." He adds that users who installed Urban VPN prior to July 2025 would have never seen the consent prompt, which was added via a silent update with version 5.5.0.

He also argues that the software provides no indication that data collection happens even when the VPN is not active.

Dardikman notes that Urban VPN received a Featured Badge from the Chrome Web Store team.

"This means a human at Google reviewed Urban VPN Proxy and concluded it met their standards," he said. "Either the review didn't examine the code that harvests conversations from Google's own AI product (Gemini), or it did and didn't consider this a problem."

He observes that the Chrome Web Store policies explicitly prohibit transferring or selling user data to third party data brokers like BiScience.

Google did not immediately respond to a request for comment.

The problem appears to be a loophole in Google's Chrome Web Store Limited Use policy, which allows data to be transferred to third parties for limited scenarios (e.g., security or business ownership change) that do not include transferring data to data brokers.

Palant in his post suggests that BiScience and its affiliated partners implement user-facing features that allegedly require access to browsing history, to claim the "necessary to providing or improving your single purpose" exception that allows limited data transfer to third parties. Or they claim the security exception by implementing safe browsing or ad blocking features.

"Chrome Web Store appears to interpret their policies as allowing the transfer of user data, if extensions claim Limited Use exceptions through their privacy policy or other user disclosures," Palant wrote. "Unfortunately, bad actors falsely claim these exceptions to sell user data to third parties."

"If you have any of these extensions installed, uninstall them now," Dardikman concluded. "Assume any AI conversations you've had since July 2025 have been captured and shared with third parties." ®