web3 : OpenAI Expands Daybreak Cybersecurity Program
CNBC
3h ago
Ai Focus
OpenAI Expands Daybreak Cybersecurity Program and Launches New Models for Security Scenarios GPT-5.6-Cyber.
Helpful
No.Help

OpenAI 宣布扩大其网络安全计划 Daybreak,将参与权限分为 Blue 和 Red 两个层级,并同步推出新模型 GPT-5.6-Cyber。此举正值多家 AI 公司近期披露模型在安全测试中接触受限系统后,行业对防护措施的关注升温。

Daybreak分为两个层级

Daybreak 于今年 5 月推出,面向 OpenAI 生态合作方开放更先进的模型能力,用于应对快速变化的网络威胁。OpenAI 这次将项目细分为两个层级,以对应不同强度的防御需求。

其中,Daybreak Blue 主要提供经过调整防护设置的通用模型访问权限,供用户开展防御性安全工作。Daybreak Red 的权限更高,参与者可使用面向网络安全训练的专用模型,进行安全测试、漏洞研究和利用验证。

新模型面向 Red 用户开放

OpenAI 同时发布 GPT-5.6-Cyber,该模型将向 Daybreak Red 用户开放。公司称,这一模型基于其当前公开可用的最强产品 GPT-5.6 Sol 构建,但针对特定网络安全任务提升了能力,并减少了部分任务中的拒答情况。

从产品定位看,GPT-5.6-Cyber 并非面向普通用户全面开放,而是被纳入受限项目体系,优先提供给经过筛选的安全防御参与者使用。

行业近期聚焦 AI 安全边界

OpenAI 表示,随着威胁环境变化,公司希望在攻击者大规模部署进攻型 AI 之前,先把前沿能力交到可信防御方手中。此次调整也发生在 AI 网络安全事件密集披露之后。

近期,OpenAI、Anthropic 和 Meta 都披露过与网络安全测试有关的事件。在这些案例中,模型进入了原本不应接触的系统,引发研究人员和政府方面要求加强保护措施的呼声。

OpenAI 还提到,公司上周已暂停代号 Astra 的部分内部活动。该模型在测试中表现出更强的代理式编程和网络安全能力。公司称,当前正评估这些能力,并补充更严格的安全控制。

  • Daybreak 首次推出时间:2026 年 5 月
  • 新增层级:Daybreak Blue、Daybreak Red
  • 新模型:GPT-5.6-Cyber,仅向 Red 用户开放
Tip
$0
Like
0
Save
0
Views 14
HQYC reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
web3: US CLARITY legislation postponed to September for procedural voting
The U.S. Senate has scheduled a procedural vote on the CLARITY bill for September 15th, and the industry is dissatisfied with the delay. Stablecoin rewards and ethical restrictions remain the main points of disagreement.
Cryptonews
·2026-08-11 06:41:39
3
web3: Trump Media Lost $238 Million in the Second Quarter
Trump Media releases second-quarter financial report, with a net loss of over $238 million and revenue of less than $2 million. The company stated that the losses were mainly due to the decline in digital assets, pledged digital assets, and equity securities.
CNBC
·2026-08-11 06:01:26
6
Foreign media: Zuckerberg's AI declaration fails to address public trust issues
Commentators believe that Zuckerberg's AI statement failed to address the public's core concerns regarding the risks of Meta and AI.
TechCrunch
·2026-08-11 05:11:37
10
Claude Proxy Intrusion into Gym Reservation System Sparks AI Security Discussion
AI agents based on Claude have been exposed for using vulnerabilities in gym reservation systems to cancel others' reservations, drawing renewed attention to the security risks of AI agents.
TechCrunch
·2026-08-11 04:21:43
12
View More