AI动力浏览器很容易受到隐藏的文字攻击,这种攻击把AI诱使它泄露数据或绕过传统安全进入恶意网站。 AI-powered browsers are vulnerable to hidden text attacks that trick AI into leaking data or visiting malicious sites, bypassing traditional security.

以AI为动力的网络浏览器面临因迅速注射袭击而增加的安全风险,网页或图像中隐藏的文字诱使AI泄露数据或访问恶意网站,而用户不知情。 AI-powered web browsers face rising security risks from prompt injection attacks, where hidden text in web pages or images tricks AI into leaking data or visiting malicious sites without user knowledge. 研究人员和勇敢的软件显示,图象中的微弱文字-如黄色的浅蓝色-命令AI浏览器,如Perplicity的彗星和Fellou进行有害行动,即使用户看不出有任何可疑之处。 Researchers and Brave Software demonstrated that faint text in images—like light blue on yellow—can command AI browsers such as Perplexity’s Comet and Fellou to perform harmful actions, even when users see nothing suspicious. 与传统的恶意软件不同,这些攻击绕过抗病毒软件,因为设备上没有代码运行。 Unlike traditional malware, these attacks bypass antivirus software since no code runs on the device. 虽然 Perplexity 在一份报告后修补了这一缺陷，但它批评 Brave 的公开警告是危言耸听。 While Perplexity patched the flaw after a report, it criticized Brave’s public warning as alarmist. 专家们敦促用户在浏览器制造者竞相解决AI驱动浏览中的系统性脆弱性时,禁用AI功能并不断更新系统。 Experts urge users to disable AI features and keep systems updated, as browser makers race to address systemic vulnerabilities in AI-driven browsing.