Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now
https://www.theverge.com/tech/1002980/google-gemini-4-argon📌 Gemini 4 Argon 發布，卻先鎖給「網路防禦者」
TL;DR：Gemini 4 Argon 效能達前沿水準，卻先限量開放給網路防禦者。
這幾天 X 上一直在傳 Gemini 4 的「洩漏版」benchmark，工程師社群議論紛紛。沒想到 Google 這次選擇正面迎戰，直接發布新模型，卻又同時宣布：多數人現在還碰不到它。
🤔 一次意外提前的正式發布
Google 今日發表新一代前沿模型 Gemini 4 Argon。根據 Google DeepMind SVP、首席 AI 架構師 Koray Kavukcuoglu 的說法，這個模型在「複雜工作流程」上達到前沿水準，範圍涵蓋真實世界的軟體工程、企業知識工作（如法律、財務），以及網路安全防禦。這次發布的時間點也頗微妙：就在 OpenAI 高調的 DevDay 隔一天，OpenAI 才剛發表 Dots AI agent 與 GPT-6.1 Sol，同時也宣布因安全疑慮而取消原訂的 GPT-6.1 Astra。而 Kavukcuoglu 本人也是今年八月才剛接任 DeepMind 掌門人。
🧩 先給「受信任的網路防禦者」，再談普及
儘管 Google 秀出一張圖表，宣稱 Gemini 4 在多項 benchmark 上超越 OpenAI 與 Anthropic 的競品模型，但實際存取權限卻相當克制。Kavukcuoglu 表示，Gemini 4 Argon 目前僅開放給「一組受信任的網路防禦者」，同時 Google 正「積極參與美國政府針對模型預發布存取的自願性流程」，並會逐步擴大開放範圍。換句話說，這不是一次面向所有開發者的公開發布，而是先鎖定在資安防禦這個高風險、高價值的應用場景做驗證。
💡 模型已經在 Google 內部跑，但外部得等安全強化
Kavukcuoglu 也透露，Gemini 4 Argon 已經在支援 Google「內部工作流程」，其中一項具體案例是協助大規模程式碼庫遷移（codebase migration）。不過在更大範圍推出之前，Google 表示要先強化「關鍵前沿防護措施」，包括防範模型被濫用、抵禦 prompt injection 攻擊，以及監控模型是否出現「misalignment」（不對齊）的跡象。這種「先驗證安全性、再擴大開放」的節奏，某種程度上呼應了業界近期對高能力模型雙重用途風險的普遍焦慮，尤其是在網路安全這種攻防皆可用的領域。
🎯 實務啟示
對工程師來說，這次發布提醒了一件事：模型能力越強，離「隨插即用」反而可能越遠。若你的團隊涉及資安防禦相關場景，值得關注 Google 後續開放存取的節奏與申請管道；若只是想比較效能，目前也只能先看官方釋出的 benchmark 圖表，實際上手測試恐怕還要一段時間。
🔗 來源
- 標題：Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now
- 作者／機構：Jay Peters, The Verge
- 連結：https://www.theverge.com/tech/1002980/google-gemini-4-argon
#Gemini4 #GoogleDeepMind #AIsafety #Cybersecurity #FrontierModel #PromptInjection #AIalignment #LLM #AIagent #TechNews
原始資料 The Verge AI · 收集於 2026-09-30
摘要原文
Tech AI News Google announces Gemini 4 and says it’s so capable that only ‘trusted cyber defenders’ can have it right now Gemini 4 Argon is launching first in a limited capacity so Google can make sure it’s not misaligned. Gemini 4 Argon is launching first in a limited capacity so Google can make sure it’s not misaligned. by Jay Peters Sep 30, 2026, 8:41 PM UTC Illustration: The Verge Jay Peters is a senior reporter covering technology, gaming, and more. He joined The Verge in 2019 after nearly two years at Techmeme. Google today revealed its next AI frontier model, which it’s calling Gemini 4 Argon. The new model delivers “frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense,” according to chief AI architect and Google DeepMind SVP Koray Kavukcuoglu. But the company is limiting access at first to a “set of trusted cyber defenders,” and Kavukcuoglu says that Google is “actively engaged in the U.S. government’s voluntary process for pre-release model access while we gradually expand access.” Kavukcuoglu says Gemini 4 Argon is already powering Google’s “internal workflows,” highlighting how the model has helped with tasks like large-scale codebase migrations. Google also shared a big chart showing many benchmarks where Gemini 4 fares better than competing models from OpenAI and Anthropic. (The AI community was buzzing earlier this week on X over what appeared to be leaked benchmarks for the new model.) Google will be strengthening “critical frontier safeguards” before a broader rollout, including defending against misuse and prompt injection attacks and monitoring for misalignment. The launch of Gemini 4 Argon comes just a day after OpenAI’s splashy DevDay conference , where OpenAI launched its Dots AI agent and GPT-6.1 Sol AI model. (OpenAI also said this week that it wouldn’t be releasing a planned GPT-6.1 Astra model due to worries about safety .) It’s also arriving soon after Kavukcuoglu’s appointment as the new boss of DeepMind in August . Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates. Jay Peters AI Google News Tech
由 claude-code/sonnet 自動生成