Daybreak 據報是 OpenAI 2026 年 5 月推出的網絡防禦平台,主打把漏洞檢查放入開發流程;Glasswing 則是 Anthropic 官方 2026 年 4 月受限計劃,向指定關鍵軟件防守方開放 Claude Mythos Preview。[3][5][13][39] OpenAI 走較可擴展但仍要審核的 Trusted Access for Cyber 路線,稱會擴至數千名已驗證防守者及數百個團隊;Anthropic 則把 Mythos Preview 限定予少數合作夥伴作防禦用途。[25][39] 兩者都唔敢全面公開,因為漏洞發掘是典型雙用途能力:可以幫開發者補洞,亦可能令攻擊者更快搵到可利用弱點。[8...

Create a landscape editorial hero image for this Studio Global article: What is OpenAI’s new Daybreak cybersecurity platform, and how does it compete with Anthropic’s Project Glasswing?. Article summary: OpenAI’s Daybreak is a reported cybersecurity initiative/platform that uses OpenAI’s GPT-5.5-era cyber capabilities to help vetted defenders find, triage, and fix software vulnerabilities, positioning it as a direct answ. Topic tags: general, general web, user generated, documentation. Reference image context from search candidates: Reference image 1: visual subject "Amazon, Microsoft and Apple gain early access to Anthropic’s latest AI model as cybersecurity becomes a frontline priority. Anthropic has launched a new AI cybersecurity initiative" source context "Anthropic Opens New AI Cybersecurity Model to Big Tech Firms Under Project Glasswing - ITP.net" Reference image 2: visual subjec
AI 網絡安全競賽,表面上係 OpenAI 對 Anthropic;但真正問題唔係邊個聊天機械人答得叻啲,而係邊間公司可以安全地將「搵漏洞」呢類高風險能力交到可信防守者手上。
Daybreak 據 MacRumors、Engadget 等媒體報道,是 OpenAI 於 2026 年 5 月推出/啟動的網絡防禦平台,利用 OpenAI 模型協助發現及修補軟件漏洞;Anthropic 的 Project Glasswing 則有較完整官方文件,是 2026 年 4 月公布、向指定關鍵軟件防守方提供 Claude Mythos Preview 的受限計劃。[3][
5][
13][
39]
換句話講,這場競爭唔係消費級 AI 產品戰,而係 AI 模型開始進入軟件安全最敏感的位置:在攻擊者之前搵出漏洞。[8][
17][
33]
Studio Global AI
Use this topic as a starting point for a fresh source-backed answer, then compare citations before you share it.
Daybreak 據報是 OpenAI 2026 年 5 月推出的網絡防禦平台,主打把漏洞檢查放入開發流程;Glasswing 則是 Anthropic 官方 2026 年 4 月受限計劃,向指定關鍵軟件防守方開放 Claude Mythos Preview。[3][5][13][39]
Daybreak 據報是 OpenAI 2026 年 5 月推出的網絡防禦平台,主打把漏洞檢查放入開發流程;Glasswing 則是 Anthropic 官方 2026 年 4 月受限計劃,向指定關鍵軟件防守方開放 Claude Mythos Preview。[3][5][13][39] OpenAI 走較可擴展但仍要審核的 Trusted Access for Cyber 路線,稱會擴至數千名已驗證防守者及數百個團隊;Anthropic 則把 Mythos Preview 限定予少數合作夥伴作防禦用途。[25][39]
兩者都唔敢全面公開,因為漏洞發掘是典型雙用途能力:可以幫開發者補洞,亦可能令攻擊者更快搵到可利用弱點。[8][17][33]
繼續“歐洲奢侈品股點解急跌,仲拖低埋大市?”以獲得另一個角度和額外的引用。
開啟相關頁面對照「Round8 招「AI Creator」點解惹嬲《Lies of P》玩家?」交叉檢查此答案。
開啟相關頁面Daybreak is OpenAI's response to Anthropic's Claude Mythos FotoField/Shutterstock OpenAI has just launched Daybreak, a cybersecurity initiative that's clearly the company's competitor to Anthropic's Project Glasswing. If you'll recall, Glasswing uses Anthro...
OpenAI introduces GPT‑5.5‑Cyber following uproar around Anthropic Mythos It's a modest upgrade focused on permissive cybersecurity tasks like vuln triage and malware analysis Access is limited to vetted teams in the Trusted Access for Cyber (TAC) program, u...
OpenAI's New Daybreak Platform Uses GPT-5.5 to Find Software Vulnerabilities OpenAI today launched Daybreak, an answer to Anthropic's Project Glasswing initiative and Mythos AI model. Like Glasswing, Daybreak is a cyber defense effort that will help tech co...
IT之家 5 月 12 日消息,OpenAI 今天(5 月 12 日)对标 Anthropic 的 Glasswing 方案,面向企业推出软件安全防御项目 Daybreak,主打把安全能力整合到软件开发阶段。 该项目主要亮点聚焦在把网络防御前置到开发流程、用 AI 自动发现高风险漏洞、并为企业提供评估与修复支持 3 个方面。 IT之家援引博文介绍,OpenAI 公布了该项目 3 个可用模型:OpenAI 公司指出,该项目并不只是一次性的漏洞扫描服务,而是一套嵌入日常开发环节的安全工具链,让软件从一开始就带上防...
簡單分,Daybreak 看起來較似 OpenAI 把 AI 網絡安全能力「產品化」成一套開發流程工具;Glasswing 則較似 Anthropic 用受控合作夥伴計劃,把 Claude Mythos Preview 投放到關鍵軟件防禦場景。[5][
6][
13][
39]
| 比較位 | OpenAI Daybreak | Anthropic Project Glasswing |
|---|---|---|
| 基本形態 | 據報是面向軟件開發流程的 AI 輔助安全平台/工具鏈。[ | Anthropic 官方文件列明的受限合作夥伴計劃。[ |
| AI 核心 | OpenAI 模型;報道提及 GPT-5.4-Cyber、GPT-5.5-Cyber 及 Codex 相關安全工具。[ | Claude Mythos Preview。[ |
| 存取模式 | 受閘口限制;OpenAI 稱 Trusted Access for Cyber 會擴至數千名已驗證個人防守者及數百個負責關鍵軟件的團隊。[ | 僅向有限合作夥伴開放,並限於防禦性網絡安全用途。[ |
| 工作流程重點 | 安全程式碼審查、威脅建模、補丁驗證、依賴風險分析、偵測及修復建議。[ | 協助保護關鍵軟件和基礎設施,尋找及修補漏洞。[ |
| 較適合邊類團隊 | 想把安全檢查嵌入日常開發流程的企業、開源安全團隊或防守者,但要視乎審批。[ | 負責關鍵軟件或基礎設施、並符合 Anthropic 夥伴准入模式的組織。[ |
Daybreak 可以理解為圍繞 OpenAI 模型建立的防禦性安全工作流,而唔係一個人人即用的消費產品。據報,它的目標是幫公司更早在開發階段發現漏洞,而唔係等到產品發布前先做最後一輪安全審查。[3][
5][
6]
報道提到 Daybreak 的應用包括安全程式碼審查、威脅建模、補丁驗證、依賴風險分析、偵測,以及修復建議。[6] MacRumors 又指 Daybreak 建基於 OpenAI 的 GPT-5.4-Cyber 工作,而 OpenAI 曾稱 GPT-5.4-Cyber 已協助修復超過 3,000 個漏洞。[
5] TechRadar 另報道,GPT-5.5-Cyber 的存取限於 OpenAI Trusted Access for Cyber 計劃內已審核團隊,重點是漏洞分流、惡意程式分析等防禦任務。[
4]
不過要講清楚:在現有來源入面,Daybreak 這個品牌本身主要來自媒體報道;OpenAI 官方資料較直接描述的是 Trusted Access for Cyber、GPT-5.5 安全工作,以及 Codex 的網絡安全防護,而唔係一個 Daybreak 專頁。[3][
5][
17][
18][
22][
25]
Project Glasswing 方面,Anthropic 的官方描述較直接:它是一個為 AI 時代保護全球最關鍵軟件的計劃,合作對象是負責大量人每天依賴之基礎設施的組織。[13]
Anthropic 列出的啟動合作夥伴包括 Amazon Web Services、Apple、Broadcom、Cisco、CrowdStrike、Google、JPMorganChase、Linux Foundation、Microsoft、NVIDIA 及 Palo Alto Networks。[13] Glasswing 的核心是 Claude Mythos Preview;Anthropic 的紅隊測試網站形容它是一個通用模型,在多項任務表現強,尤其擅長電腦安全工作。[
33]
Anthropic 透明度中心亦列明,Claude Mythos Preview 只會作為 Project Glasswing 一部分,向有限合作夥伴開放作防禦性網絡安全用途,發布日期列為 2026 年 4 月。[39] Anthropic 亦曾表示會維持 Mythos Preview 的有限發布,並先在能力較低的模型上測試新的網絡安全防護措施。[
34]
其中一個公開案例來自 Engadget 報道:Mozilla 表示 Mythos 協助其在最新 Firefox 版本中發現並修補 271 個漏洞。[3]
OpenAI 同 Anthropic 的競爭,唔只係模型能力。更關鍵係:當 AI 已經可以協助找出高價值漏洞,平台要點樣判斷邊個可以用、用到幾深、出事點樣截停?
OpenAI 的路線較似「擴大但受控」。它把 Trusted Access for Cyber 描述為一套讓先進網絡能力廣泛交到防守者手上,但存取會隨身份、信任、驗證及防護措施而調整的框架。[18][
29] OpenAI 又稱,正把計劃擴展至數千名已驗證個人防守者,以及數百個負責防禦關鍵軟件的團隊。[
25]
Anthropic 的路線則更明顯集中於「少量但高信任」的合作夥伴。Glasswing 讓指定防守者使用 Claude Mythos Preview,而 Anthropic 自己的資料亦強調,由於 Mythos Preview 在網絡安全方面能力突出,所以未作更廣泛發布。[33][
34][
39]
漏洞發掘是雙用途能力。今日它可以幫維護者找出記憶體安全問題、驗證補丁是否有效;但同一種能力,如果落入攻擊者手上,也可以幫他們更快鎖定可利用弱點。TechXplore 形容 OpenAI 和 Anthropic 的受限發布,反映業界擔心 AI 令防守者與黑客之間出現軍備競賽。[8]
OpenAI 的安全文件亦反映這種取態。GPT-5.5 system card 指,模型在發布前經過針對高階網絡安全及生物能力的紅隊測試,並配備旨在減少濫用、同時保留正當用途的防護措施。[22] GPT-5.5 Instant system card 則表示,這是 OpenAI 首個在網絡安全及生物與化學準備分類中,被視為 High capability 的 Instant 模型。[
19] 至於 Codex,OpenAI 稱系統可拒絕明顯惡意請求,並透過分類器監測可疑網絡活動,把高風險流量轉到網絡能力較低的模型。[
17]
第一,睇准入條件。OpenAI 說 Trusted Access for Cyber 會擴大至數千名已驗證防守者和數百個團隊,但仍然要經過審核及安全機制。[25] Anthropic 目前公開立場則更保守,Mythos Preview 只向有限合作夥伴作防禦用途開放。[
39]
第二,睇證據質素。公開「修了幾多漏洞」當然吸睛,但保安團隊更需要清楚案例:究竟找到的是甚麼類型漏洞?補丁是否正確?工具點樣融入現有程式碼審查、測試、事故應變流程?在現有來源入面,較具體的 Glasswing 相關公開例子,是 Mozilla/Firefox 涉及 271 個漏洞的報道。[3]
第三,睇安全機制。OpenAI 偏向用身份及信任基礎准入、拒絕惡意請求、監測、流量轉路由及模型層防護。[17][
18][
22] Anthropic 則偏向有限發布、合作夥伴准入,以及先在能力較低模型上測試新防護措施。[
34][
39]
OpenAI Daybreak 看起來是 OpenAI 對 Anthropic Project Glasswing 的產品化回應,但兩者公開文件並不對稱。Glasswing 有 Anthropic 官方頁面和透明度資料;Daybreak 在現有來源中主要由媒體報道支撐,旁邊再配合 OpenAI 官方的 Trusted Access 及模型安全資料。[3][
5][
13][
18][
25][
39]
實際 takeaway 好清楚:高階 AI 漏洞發掘正由研究示範,走入需要審批、身份驗證及安全防護的防守者工作流。Daybreak 目前最突出的賣點,是把 AI 安全能力放入日常開發;Glasswing 最突出的賣點,則是把 Claude Mythos Preview 受控開放予保護關鍵軟件的組織。[5][
6][
13][
39]
OpenAI announces restricted-access cybersecurity model ... Artificial intelligence company OpenAI said Tuesday that it would release its latest cybersecurity model to a limited number of partners, after rival Anthropic also restricted release of a new syste...
What is Project Glasswing Project Glasswing is an initiative to secure the world’s most critical software for the AI era. We're partnering with the organizations responsible for the infrastructure billions of people depend on, and giving their defenders a h...
GPT-5.3-Codex is the first model we are treating as High cybersecurity capability under our Preparedness Framework, which requires additional safeguards. These safeguards include training the model to refuse clearly malicious requests like stealing credenti...
Trusted Access for Cyber is designed around a simple premise: advanced cyber capabilities should reach defenders broadly, but access should scale with trust, validation, and safeguards. Today we’re sharing the first organizations helping put that approach...
GPT-5.5 Instant is our latest Instant model, and explained in our blog. The comprehensive safety mitigation approach for this model is similar to previous models in this series, but this is the first Instant model that we are treating as High capability in...
April 23, 2026 ... We subjected the model to our full suite of predeployment safety evaluations and our Preparedness Framework, including targeted red-teaming for advanced cybersecurity and biology capabilities, and collected feedback on real use cases from...
We continue to evolve trusted access, safeguards, and ecosystem support to help cyber defenders protect us all. ... We are scaling up our Trusted Access for Cyber (TAC) program to thousands of verified individual defenders and hundreds of teams responsible...
GPT‑5.3-Codex is our most cyber-capable frontier reasoning model to date. ... To unlock the full defensive potential of these capabilities while reducing the risk of misuse, we are piloting Trusted Access for Cyber: an identity and trust-based framework des...
Earlier today we announced Claude Mythos Preview, a new general-purpose language model. This model performs strongly across the board, but it is strikingly capable at computer security tasks. In response, we have launched Project Glasswing, an effort to use...
Introducing Claude Opus 4.7 Apr 16, 2026 Our latest model, Claude Opus 4.7, is now generally available. Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks. ... Last week we annou...
Model description Claude Mythos Preview is a general-purpose frontier model with advanced agentic coding and reasoning skills. It is being made available to a limited set of partners for defensive cybersecurity purposes only, as part of Project Glasswing. ....