跳到主要內容

XBOW評測:GPT-5.6挖出Chrome零日,奪25萬美元賞金 | XBOW Review: GPT-5.6 Finds Chrome Zero-Day, Wins $250K

By Kit 小克 | AI Tool Observer | 2026-10-15

🇹🇼 XBOW評測:GPT-5.6挖出Chrome零日,奪25萬美元賞金

XBOW用GPT-5.6挖出Chrome零日漏洞,抱回25萬美元賞金

資安研究公司 XBOW 近日公開一項測試結果:他們用一套完全公開、人人都能租用的 AI 模型 GPT-5.6 Sol,在六天內花費約 4 萬美元運算成本,串連三個漏洞做出完整的 Chrome 零日攻擊鏈,成功從瀏覽器分頁的 JavaScript 一路突破渲染沙箱、執行任意程式碼。Google 隨後支付了修訂版漏洞賞金計畫(VRP)的最高金額:25 萬美元。

三個漏洞怎麼串起來的

這條攻擊鏈踩中三個不同元件:先是 V8 JavaScript 引擎裡 Maglev 編譯器的錯誤;接著是動畫 WebP 圖片解碼器的瑕疵(編號 CVE-2026-85045);最後靠 Crashpad(Chrome 的錯誤回報機制)裡的記憶體安全問題完成逃出沙箱。這類跨元件串接過去通常要頂尖資安研究員耗時數週甚至數月才拼得出來,XBOW 的團隊讓 AI 在六天內就走完大部分流程,活動期間落在 2026 年 8 月 2 日到 7 日,10 月 13 日才對外公開。

為什麼刻意選一般人都能用的模型

XBOW 沒有用內部客製化或限制存取的模型,而是選任何開發者都能透過 API 呼叫的 GPT-5.6 Sol,目的是證明:挖掘高價值 Chrome 零日的能力,已經不是少數實驗室的專利。報告也特別強調,人類研究員仍負責關鍵判斷——AI 負責大量嘗試與分析,但往哪個方向切入、何時該收斂,還是靠人拍板。這起漏洞原本可能影響約 10 億名 Windows 版 Chrome 用戶,Google 已在公開前完成修補,使用者更新瀏覽器即可免除風險。

對開發者與資安團隊的意義

對一般使用者來說,這次事件的重點不是恐慌,而是提醒:保持瀏覽器自動更新,三個漏洞都已在你我更新 Chrome 時悄悄補上。對資安與 AI 從業者而言,GPT-5.6 挖出 Chrome 零日的案例是具體證據——自動化攻防的門檻正快速下降,防守方也該開始用同一套工具檢查自己的程式碼,而不是等漏洞被別人先找到。

好不好用,試了才知道。


🇺🇸 XBOW Review: GPT-5.6 Finds Chrome Zero-Day, Wins $250K

XBOW Used GPT-5.6 to Find a Chrome Zero-Day, Won $250K

Security research firm XBOW disclosed that it chained three bugs into a full Chrome zero-day exploit using nothing but GPT-5.6 Sol — a model anyone can rent through OpenAI's public API. The campaign ran for six days and cost about $40,000 in compute, escaping Chrome's renderer sandbox entirely, from JavaScript in a browser tab to arbitrary code execution. Google paid the new maximum payout under its revised Chrome Vulnerability Reward Program: $250,000.

How the Three Bugs Chained Together

The exploit touched three separate components: a bug in the Maglev compiler inside V8, Chrome's JavaScript engine; a flaw in the animated WebP image decoder, tracked as CVE-2026-85045; and a memory-safety issue in Crashpad, Chrome's crash-reporting tool. Chaining bugs like this across unrelated subsystems usually takes elite human researchers weeks or months. XBOW's AI-driven pipeline got most of the way there in six days, running from August 2–7, 2026, and going public on October 13.

Why a Public Model, Not a Custom One

XBOW deliberately avoided a specialized or access-restricted model, picking GPT-5.6 Sol — the same model any developer can call today — to make a point: finding high-value bugs in software like Chrome is no longer locked inside a handful of elite labs. Still, the team is clear that humans made the critical calls; the AI handled brute-force exploration and analysis, while direction and judgment stayed human. The flaw chain could have affected roughly 1 billion Chrome users on Windows. Google patched all three bugs before the disclosure went public, so a routine browser update already covers you.

What It Means for Developers and Security Teams

For everyday users, this isn't a reason to panic — just a reminder to keep auto-updates on. For developers and security teams, the GPT-5.6 Chrome zero-day is concrete evidence that the cost of automated offensive security research is dropping fast, which means defenders should start pointing the same kind of tools at their own codebases before someone else does.

You won't know if it's worth it until you try it.

Sources / 資料來源

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code