Вредоносное расширение может перехватить контроль над ИИ-помощниками в Chromium-браузерах

Security researcher Gal Weizman of Forever Security has disclosed BragJack, a proof-of-concept attack technique letting a malicious browser extension seize control of built-in AI assistants in Chromium-based browsers. The method, which Weizman calls "Prompt Forcing," abuses an extension's normal page-modification and declarativeNetRequest permissions to inject code into the trusted web context that governs the AI agent, then feeds it attacker-crafted commands the agent treats as legitimate instructions. Affected products include Gemini Live in Google Chrome, Perplexity Comet, Microsoft Edge, Opera Neon and Claude for Chrome; impacts range from reading local files and taking screenshots to exfiltrating email contents, as shown in Comet. Google tracked its flaw as CVE-2026-0628 (CVSS 8.8) and fixed it in Chrome 143.0.7499.192, while Microsoft assigned CVE-2026-55945 and patched it in Edge 150.0.4078.48; Comet, Opera Neon and Claude for Chrome issues remain without CVE IDs so far. Weizman received a combined $20,500 in bug bounty rewards, and no real-world exploitation has been observed, though the technique still requires tricking a victim into installing the malicious extension.