返回全部动态

Cerebras:网络防御为何需要更快的推理速度

原标题:Why Cyber Defense Needs Faster Inference September 24, 2026

Cerebras Blog一手来源观点质量 10

AI 摘要

Cerebras 博客文章指出,AI 智能体正以更快的速度和规模发动网络攻击,CrowdStrike 报告显示 AI 驱动的攻击操作同比增长 89%,最快突破时间仅 27 秒。OpenAI 的 Greg Brockman 透露已将 25% 的生产工程师调去构建「防御工厂」。Cerebras 借此宣传其推理速度优势,称运行 GPT-5.6 Sol Ultrafast 比标准处理快 14 倍,在恶意文件分析任务中耗时 58 秒,而标准处理需 4 分 43 秒,并已与 CrowdStrike Falcon 平台合作。

以上摘要由 AI 生成,可能存在误差。事实请以原文为准。

正文节选

In recent months, increasingly sophisticated AI agents have infiltrated real production infrastructure. One gained code execution access on dozens of a third party’s servers. Another uploaded malicious software and accessed a live database. A third breached three companies using guessed or exposed credentials. These incidents show how AI is transforming cybersecurity: swarms of AI agents are powering attacks at great speed and scale, and defenders must continually outpace them. The urgency of cy


发布时间:—
抓取时间:2026-09-24 23:36
来源机构:Cerebras
阅读原文cerebras.ai