① 他说过的话

15 条

它基本上就是我喜欢称之为「恶意软件入门套装」的东西，因为如果你心情慷慨或处于混乱邪恶阵营，你几乎会原封不动把这些交给一段恶意软件。
it’s basically what I like to call the Malware Starter Pack, because it’s almost exactly what you would hand a piece of malware if you were feeling generous or chaotic evil.
—— Sarah Sanders · [04:00]

指向原始笔记的链接

而提示词不是安全。
And prompts are not security.
—— Sarah Sanders · [05:47]

指向原始笔记的链接

我学到的教训是：攻击会组合，而代码审查不会，因为我们开发者一次只看一个 diff,但攻击者看的是整个系统，他们会寻找那两个握手并打开一扇门的东西。
So the lesson I learned was that attacks compose, code review doesn’t, because us developers all look at diffs one at a time, but attackers look at the whole system and they look for those two things that shake hands and open a door.
—— Sarah Sanders · [08:10]

指向原始笔记的链接

正是这个威胁重塑了我对安全以及 wizard 的思考，因为对我们来说，危险输入真的可能来自我们自己的供应链。
So that was the threat that reshaped how I think about security and the wizard, because the dangerous input for us really could come from our own supply chain.
—— Sarah Sanders · [10:16]

指向原始笔记的链接

我的方法论是：在源头抓住它，假设源头失败了，再在使用点抓住它一次。
My methodology is catch it at the source, assume the source failed, and catch it again at the point of use.
—— Sarah Sanders · [10:38]

指向原始笔记的链接

然后「大概没问题」就不再够好了。
And where probably fine just stops being good enough.
—— Sarah Sanders · [11:20]

指向原始笔记的链接

因为检测问题是一份工作，而决定如何处理那个问题是完全不同的另一份工作。
Because detecting a problem is one job, and deciding what to do about that problem is a totally different job.
—— Sarah Sanders · [12:05]

指向原始笔记的链接

它故意做得无聊，而在安全领域，无聊是一个特性。
It’s boring on purpose, and in security, boring is a feature.
—— Sarah Sanders · [12:32]

指向原始笔记的链接

那个机器人有一个要完成的任务，它在试图优化并取悦我们，但我们不能允许这样。
The robot had a task to do and it was trying to optimize and please us, but we can’t have that.
—— Sarah Sanders · [13:29]

指向原始笔记的链接

智能体真的不在乎暴露数据，除非你制定明确的规则。
Agents genuinely do not care about exposing data unless you make explicit rules.
—— Sarah Sanders · [13:40]

指向原始笔记的链接

虽然这很诱人，因为看起来更简单，但我不能把我的安全模型押在掷硬币上，就因为我的模型今天状态不好、行为和昨天不一样。
And while that’s tempting because it seems easier, I can’t bet my security model on a coin flip because my model’s having a bad day or something happened and it’s acting different today than it did yesterday.
—— Sarah Sanders · [15:14]

指向原始笔记的链接

如果命中一条规则，闸门锁上，会话结束，这条路径上没有任何模型参与。
If a rule matches, the gate locks, the session ends, and there is no model anywhere on that path.
—— Sarah Sanders · [15:44]

指向原始笔记的链接

执行是你押上身家性命的部分，所以它必须是确定性的；而判断是增加细微差别的部分，那才是你唯一能放进概率性东西的地方。
Enforcement is the part that you bet the house on, so it has to be deterministic, but judgment is the part that adds nuance, so that’s really the only place that you can put anything probabilistic in there.
—— Sarah Sanders · [16:15]

指向原始笔记的链接

你要为你正在构建的那个智能体判断真实世界的影响，因为一个每次试图清理构建文件夹就崩溃的安全工具，是一个会被关掉的、什么也抓不到的工具。
for the agent that you’re building because a security tool that crashes every time it tries to clean a build folder is a tool that gets turned off and one that catches absolutely nothing.
—— Sarah Sanders · [18:36]

指向原始笔记的链接

第一，凡是没有被确定性强制执行的，就等于没有强制执行。
One, if it isn’t enforced deterministically, it is not enforced.
—— Sarah Sanders · [19:48]

指向原始笔记的链接

② 出现在这些集

1 集

• 《PostHog 把智能体装进终端，再给它请了个“保镖”》 — 作为嘉宾

③ 他谈到的

点进去有真内容 —— 本页主要出口

PostHog · wizard · Warlock · 智能体 · 提示词注入 · 沙箱 · 护栏 · 子智能体 · Yara · 上下文引擎

④ 也在聊「AI 安全」的人