新聞Google Cloud本周在RSA大會上宣布多項AI安全工具，包括將Gemini AI部署到暗網以提升情報精確性，以及可由代理人蒐集證據的企業安全工具等。現今企業安全團隊的工作挑戰之一在於接收到大量不準確的資料，為了提供更精準的情報給企業客戶，Google從二個方面整合AI代理人和其Google威脅情報（Google Threat Intelligence）。其中之一是由內建最新Gemini…
新聞3月17日圖像化大型語言模型（LLM）開發工具Langflow開發工程團隊發布資安公告，指出此工具存在未經身分驗證的遠端程式碼執行（RCE）漏洞CVE-2026-33017，該漏洞發生的原因在於，特定端點允許任何人在未經身分驗證的狀態下建立公開的工作流程，一旦攻擊者提供選用參數，就可能導致端點使用攻擊者控制的流程資料，而非Langflow資料庫存放的資料。相關的程式碼因此不會通過沙箱處理，並導…
新聞上週Langflow開發工程團隊揭露與修補重大風險漏洞CVE-2026-33017，通報此事的研究人員Aviral Srivastava指出，該漏洞發生的原因與一年前揭露CVE-2025-3248雷同，攻擊者都能在未經身分驗證的情況下遠端執行任意程式碼。由於CVE-2025-3248公布不久後就出現攻擊行動，CVE-2026-33017也有可能成為被利用的對象，果不其然，有資安公司發現，駭客在…
新聞日本汽車大廠馬自達（Mazda）本週揭露，進口零件相關的倉儲管理系統遭駭客入侵，致使約700筆員工及合作夥伴資訊被竊取。馬自達是在2025年12月中發現該公司一個儲存自泰國採購的汽車零件的倉儲管理系統遭未授權方由外部存取。馬自達除了第一時間採取資安措施、通報主管機關和警方外，也偕同外部資安專家進行調查。本週馬自達公布調查結果，並發出道歉啟事。駭客是利用公司某個營運系統的安全漏洞，得以存取系統資…
新聞Anthropic週一（3/23）宣布，在Claude Cowork與Claude Code中推出「電腦使用」（Computer Use）功能，讓AI可直接透過滑鼠、鍵盤與螢幕操作電腦，完成開啟檔案、瀏覽網頁與執行開發工具等任務，正式將AI推向「電腦操作代理人（Computer-using Agents）」階段。Claude Cowork與Claude Code為Anthropic建立於桌面應…
新聞Anthropic上週公布Claude Code Channels功能，讓開發人員利用外部通訊工具如Telegram、Discord和Claude溝通，遠距就能下達指令。Claude Code Channels本質上是一種「事件通道（event channel）」機制，讓開發人員可以把訊息、警示或webhooks從MCP伺服器主動推送進一個正在運行的Claude Code session，而不…
‘People don’t like wearing things on their faces and don’t trust those who do’Science fiction author Neal Stephenson, who coined the term “metaverse” in his 1992 novelSnow Crash, has argued he and oth…
Trump Administration To Pay French Company $1 Billion To Stop Offshore Wind Farms https://news.slashdot.org/story/26/03/23/2221232/trump-administration-to-pay-french-company-1-billion-to-stop-offshore…
Republicans in Congress add $250 annual federal EV tax to transport bill Five principles should determine how you fund transport, says Consumer Reports. https://arstechnica.com/cars/2026/03/republican…
Apple will talk iOS 27, macOS 27, and more at WWDC 2026 on June 8 WWDC is often light on hardware, but it's where big software announcements happen. https://arstechnica.com/gadgets/2026/03/apple-will-…
Researchers say persona-based prompting can improve works for safety but not for factsMany people start their work with AI by prompting the machine to imagine it is an expert at the task they want it …
GTK3 Toolkit Winding Down To One Release Per Year The GTK 4.0 toolkit released in December 2020 while the GTK3 toolkit has continued to be maintained given a lot of software still relying on that olde…
Phoronix@masto.ai
·
'It freakin' worked' says Rob Joyce - and shows how relentless AI agents can find holes humans missRSAC 2026 The now-infamousAnthropic report about Chinese cyberspiesabusing Claude AI to automate cybe…
Nvidia CEO Says He's 'Empathetic' To DLSS 5 Concerns https://tech.slashdot.org/story/26/03/23/2214245/nvidia-ceo-says-hes-empathetic-to-dlss-5-concerns?utm_source=rss1.0mainlinkanon
A mission NASA might kill is still returning fascinating science from Jupiter "We can’t quite afford to support everything that we have done in the past." https://arstechnica.com/space/2026/03/a-missi…
Customers are 'excited' says one solution providerSnowflake is putting cash and kinetic energy behind the idea that AI works best in its platform.…
Washington content to be represented by actual empty chairsRSA 2026 Back in the day (circa 2023) when cybercrime group Scattered Spider and its help-desk voice-phishing calls were a relatively new thr…
Bipartisan Bill Seeks To Ban Sports Betting On Prediction Market Platforms https://yro.slashdot.org/story/26/03/23/2015241/bipartisan-bill-seeks-to-ban-sports-betting-on-prediction-market-platforms?ut…
Trump's MAHA pick for surgeon general flounders amid GOP doubts She stalled over MAHA woo-woo, anti-vaccine views, and lacking medical background. https://arstechnica.com/health/2026/03/trumps-maha-pi…
Here's where you ought to spend your security billable hours budget this yearStrengthen your MFA policies, double-down on anti-phishing training, and for Jobs' sake, patch all your vulns right away. T…
