Skip to main content

TrendAI™ 深度研究

spark

Featured Articles

Seven Principles Behind the Adoption of AI in Cybercrime
AI

AI 網路犯罪七大運作原則:為何駭客追求規模化,而非全自動攻擊

AI 尚未催生全新的網路犯罪模式,卻正讓既有犯罪手法變得更快速、成本更低,也更容易規模化。 趨勢科技 TrendAI™ 最新研究歸納出網路犯罪者採用 AI 的七大原則,解析 AI 如何提升網路釣魚與詐騙活動的執行效率、加速被竊資料的分析與處理,並逐步應用於惡意程式開發。同時指出,AI 代理程式雖已被用於實際入侵,但其自主攻擊能力仍受到技術可靠性、基礎設施存取權及人為監督等條件限制。

Read article
Beyond Ransomware: Tracking the Underground Trade in Law Firm Access and Data
Cybercriminal underground
不只勒索軟體:法律事務所存取權限與機密資料遭地下市場反覆轉售

法律事務所的資安風險,未必會隨勒索軟體攻擊結束而消失。趨勢科技 TrendAI™ 研究團隊發現,網路犯罪者會在地下市場反覆兜售法律事務所的系統存取權限與機密資料,部分交易甚至在資安事件公開前就已出現,事件結案後仍持續流通。研究顯示,公開的勒索軟體資料外洩紀錄,可能只是法律產業所面臨資安威脅的一部分。

Mayra Rosario Fuentes
Read article
AI governance AI
AI 代理程式(AI Agent)資安風險與防護:NVIDIA OpenShell 與 TrendAI Vision One™ 實測解析

AI 代理程式(AI Agent)不只是聊天機器人,更能自主規劃與執行任務、呼叫外部工具,甚至與其他 AI 代理程式協作。隨著自主性提高,企業也面臨提示詞注入攻擊(Prompt Injection)、權限過大及機密資訊外洩等資安風險。 趨勢科技 TrendAI™ 透過實際測試,探討如何結合 NVIDIA OpenShell 與 TrendAI Vision One™,運用最小權限原則及涵蓋預防、偵測與應變的多層安全控管機制,協助企業降低 AI 代理程式帶來的資安風險,強化安全治理。

Naonari Tanimoto (Solutions Architect Specialist, Trend AI), Writing Support: Omi Tokunaga (Solutions Architect, NVIDIA G.K.)
Read article
The Browser Is Now an AI Runtime: Security Risks in Chrome's On-Device Models
Cyber risk AI
Google Chrome 內建 AI 模型的三類資安風險:提示詞注入、運算資源濫用與工具呼叫攻擊

Google Chrome 會在符合條件的桌面裝置上自動下載約 4.27 GB 的 Gemini Nano 語言模型,使瀏覽器成為裝置端 AI 執行環境。然而,在現行架構下,任何 HTTPS 網站都能在沒有使用者同意提示、使用狀態指示及速率限制的情況下,於背景呼叫該模型,衍生提示詞注入(Prompt Injection)、運算資源濫用(Parasitic Computing),以及實驗性工具呼叫功能所涉及的潛在 Return-to-Tool(RTT)攻擊風險。 本文深入解析這三類資安風險,並提供企業資安團隊與開發者具體的防護及治理建議。

Kirill Gelfand, Vladimir Kropotov, Fyodor Yarochkin, David Tulis
Read article
Spam Campaign Hopes to Lure Royal Bank of Canada Customers

A new spam campaign uses classic tricks to lure clients of the Royal Bank of Canada.

Read Article
Cyber crime Cybercriminal underground
Data Breaches 101: How They Happen, What Gets Stolen, and Where It All Goes

Data breaches take time and a lot of effort to pull off, but successful breaches can affect not just organizations, but also millions of people. Learn what a data breach is, what types of data is usually stolen, and what happens to stolen data.

Read Article
Fortnite for Android Released Outside the Google Play Store. Is it Safe?

Fortnite: Battle Royale, Epic Games' massively popular online game, is being released on Android, but not on Google Play Store. Does it pose a security risk?

Read Article
Manufacturing IoT & smart devices
A Look Into Smart Factories: A Model of IIoT Innovation

The IIoT is a catalyst for development in many industries. For manufacturing, IIoT innovation is best demonstrated by one of its prominent models, the smart factory.

Read Article
Data Breaches Highlight the Need for Managed Detection and Response

With today’s ever-evolving threat landscape, data breaches are no longer isolated cases. Responding to and remediating data breaches calls for a proactive approach — something that managed detection and response (MDR) can provide.

Read Article
Security by Design: DevOps in the Era of the GDPR [Infographic]

How can DevOps — complemented with multilayered security solutions — align with the GDPR’s mandate of data protection by design and default?

Read Article
EU-JEPA Inked, Reciprocal Adequacy of Japan Data Protection Law and the GDPR Finalized

Japan’s reformed data privacy law is recognized as “equivalent” to the EU’s GDPR. The recognition of reciprocal adequacy creates the world’s largest zone of safe data transfers and reinforces the importance of data protection in international trade.

Read Article
Jigsaw Ransomware Resurfaces as Bitcoin Stealer

Security researchers uncovered that a version of Jigsaw, an old ransomware, has resurfaced as a bitcoin stealer. Its operators have already netted 8.4 bitcoins (US$66,807 as of July 24, 2018) using the repurposed malware.

Read Article
Is Big Data Enough for Machine Learning in Cybersecurity?

Threat data — enough of it — is critical to a machine learning system’s success in cybersecurity solutions. But is data quantity the be-all and end-all of effective machine learning?

Read Article
ESLint Submodules Hacked: Devs Advised to Enable Layered Authentication, Renew Tokens

An attacker compromised popular Javascript code library ESLint’s submodule using a valid npm maintainer’s account to steal approximately 4,500 log in credentials, which can be used to compromise other developers’ Javascript projects.

Read Article