OpenAI总裁Greg Brockman：网络安全物理法则正在改写，安全投入应远超软件本身OpenAI President Greg Brockman: The Physics of Cybersecurity Are Changing — Security Spend Should Vastly Outstrip Software Spend
The Takeaway：AI正在让软件变得极其便宜，真正的价值会转移到可靠性与安全性上，因此安全支出必须远超软件支出本身。
OpenAI联合创始人兼总裁Greg Brockman在No Priors播客中强调，当前处于“防御者窗口期”——前沿模型已能帮助防御方发现并修复漏洞，但这些能力尚未广泛扩散。他指出人类编写的代码本质上不安全，互联网是为弹性而非安全而建，现在是用AI偿还技术债的时刻。OpenAI已内部暂停部分项目，动用峰值智能模型扫描基础设施并修复问题，同时承诺10亿美元支持一线防御者。
Brockman认为：“网络支出、安全支出实际上应该远远超过软件支出。”因为软件门槛正在消失，人们真正需要的是可靠、可信任的系统。他同时谈到数学突破已证明AI能产生新知识，并将同一逻辑推向药物发现：生物数据是瓶颈，但监管文件、N-of-1疾病匹配等环节可迅速加速。最终他重申OpenAI的核心信念——人类必须牢牢坐在驾驶座上，AI应放大人类能动性而非取代它。
The Takeaway: AI is making software extremely cheap; the real value shifts to reliability and security, so security spending must vastly exceed software spending itself.
OpenAI co-founder and President Greg Brockman argues we are in a “defender’s window” where frontier models can already help organizations find and fix vulnerabilities before those capabilities diffuse widely. Human-written code is fundamentally insecure; the internet was built for resilience, not security. OpenAI has already paused projects, applied its peak models to its own infrastructure, and committed $1 billion to frontline defenders. “Cyber spend, security spend should actually vastly outstrip software spend,” he says, because once software is abundant, the only thing that matters is whether you can trust it. The same data-and-checkability logic that cracked century-old math problems is now being pointed at drug discovery, though wet-lab and regulatory bottlenecks remain. Throughout, Brockman insists humans must stay firmly in the driver’s seat.
OpenAI联合创始人兼总裁Greg Brockman在No Priors播客中强调，当前处于“防御者窗口期”——前沿模型已能帮助防御方发现并修复漏洞，但这些能力尚未广泛扩散。他指出人类编写的代码本质上不安全，互联网是为弹性而非安全而建，现在是用AI偿还技术债的时刻。OpenAI已内部暂停部分项目，动用峰值智能模型扫描基础设施并修复问题，同时承诺10亿美元支持一线防御者。
Brockman认为：“网络支出、安全支出实际上应该远远超过软件支出。”因为软件门槛正在消失，人们真正需要的是可靠、可信任的系统。他同时谈到数学突破已证明AI能产生新知识，并将同一逻辑推向药物发现：生物数据是瓶颈，但监管文件、N-of-1疾病匹配等环节可迅速加速。最终他重申OpenAI的核心信念——人类必须牢牢坐在驾驶座上，AI应放大人类能动性而非取代它。
The Takeaway: AI is making software extremely cheap; the real value shifts to reliability and security, so security spending must vastly exceed software spending itself.
OpenAI co-founder and President Greg Brockman argues we are in a “defender’s window” where frontier models can already help organizations find and fix vulnerabilities before those capabilities diffuse widely. Human-written code is fundamentally insecure; the internet was built for resilience, not security. OpenAI has already paused projects, applied its peak models to its own infrastructure, and committed $1 billion to frontline defenders. “Cyber spend, security spend should actually vastly outstrip software spend,” he says, because once software is abundant, the only thing that matters is whether you can trust it. The same data-and-checkability logic that cracked century-old math problems is now being pointed at drug discovery, though wet-lab and regulatory bottlenecks remain. Throughout, Brockman insists humans must stay firmly in the driver’s seat.
The Takeaway: AI is making software extremely cheap; the real value shifts to reliability and security, so security spending must vastly exceed software spending itself.
OpenAI co-founder and President Greg Brockman argues we are in a “defender’s window” where frontier models can already help organizations find and fix vulnerabilities before those capabilities diffuse widely. Human-written code is fundamentally insecure; the internet was built for resilience, not security. OpenAI has already paused projects, applied its peak models to its own infrastructure, and committed $1 billion to frontline defenders. “Cyber spend, security spend should actually vastly outstrip software spend,” he says, because once software is abundant, the only thing that matters is whether you can trust it. The same data-and-checkability logic that cracked century-old math problems is now being pointed at drug discovery, though wet-lab and regulatory bottlenecks remain. Throughout, Brockman insists humans must stay firmly in the driver’s seat.
查看原文 →
OpenAI co-founder and President Greg Brockman argues we are in a “defender’s window” where frontier models can already help organizations find and fix vulnerabilities before those capabilities diffuse widely. Human-written code is fundamentally insecure; the internet was built for resilience, not security. OpenAI has already paused projects, applied its peak models to its own infrastructure, and committed $1 billion to frontline defenders. “Cyber spend, security spend should actually vastly outstrip software spend,” he says, because once software is abundant, the only thing that matters is whether you can trust it. The same data-and-checkability logic that cracked century-old math problems is now being pointed at drug discovery, though wet-lab and regulatory bottlenecks remain. Throughout, Brockman insists humans must stay firmly in the driver’s seat.