--- title: "💢💢💢" type: "Topics" locale: "zh-CN" url: "https://longbridge.com/zh-CN/topics/39597029.md" description: "🔥Anthropic 意外泄露 “最强模型” 后，我开始重新看待 $Palo Alto Networks(PANW.US) $CrowdStrike(CRWD.US) $Zscaler(ZS.US) 的护城河市场今天砸的，不只是一次意外泄露。Anthropic 一款尚未正式发布的新模型 “Claude Mythos” 因为草稿文件被放进可公开访问的数据缓存而提前曝光。更刺激市场的，不只是 “最强模型” 这几个字，而是泄露内容里那句更重的话：它被描述为可能带来 “前所未有的网络安全风险”..." datetime: "2026-03-28T09:21:57.000Z" locales: - [en](https://longbridge.com/en/topics/39597029.md) - [zh-CN](https://longbridge.com/zh-CN/topics/39597029.md) - [zh-HK](https://longbridge.com/zh-HK/topics/39597029.md) author: "[辰逸](https://longbridge.com/zh-CN/profiles/16318663.md)" --- > 支持的语言: [English](https://longbridge.com/en/topics/39597029.md) | [繁體中文](https://longbridge.com/zh-HK/topics/39597029.md) # 💢💢💢 🔥After Anthropic's accidental leak of the "strongest model," I've started to re-evaluate the moats of $Palo Alto Networks(PANW.US), $CrowdStrike(CRWD.US), and $Zscaler(ZS.US). What the market hammered today wasn't just an accidental leak. Anthropic's new, not-yet-officially-released model "Claude Mythos" was prematurely exposed because draft files were placed in a publicly accessible data cache. What further spooked the market wasn't just the words "strongest model," but the heavier statement within the leaked content: it was described as potentially posing an "unprecedented cybersecurity risk." Anthropic also confirmed the model's existence, calling it a "step-function breakthrough" in capabilities, currently still in an early access testing phase. At the opening of US stocks on Friday, the cybersecurity sector was almost collectively pushed down. What I saw wasn't negative news for a single company, but the market re-evaluating a question in the most direct way: when large models themselves start to possess stronger offensive and defensive capabilities, are traditional security software companies actually facing new demand, or are they having their value chains rewritten? During the session, $Palo Alto Networks(PANW.US) fell to $145.08, $CrowdStrike(CRWD.US) to $363.48, $Zscaler(ZS.US) to $130.45, $Datadog(DDOG.US) to $115.02, $Okta(OKTA.US) to $73.96, and $CloudFlare(NET.US) also dropped to $201.93. I think the most interesting aspect of this decline isn't the overly simplistic conclusion of "will AI replace cybersecurity," but that the market has started to trade a more nuanced logic in advance: if the new generation of models can already significantly improve the efficiency of vulnerability discovery, code understanding, attack path simulation, and automated exploitation, then many moats built on manpower, rule libraries, and traditional workflows in the past will be forced to be re-evaluated. The leaked document was interpreted by outsiders as indicating this model's clear lead in cybersecurity capabilities, amplifying this concern. But I don't think this can be simply understood as "security stocks are done." The real problem is precisely more complex. The stronger AI gets, the larger the attack surface becomes, and the harder defense gets—that's the first layer. Yet the same AI capabilities will also be rapidly absorbed by security vendors, becoming new engines for detection, response, remediation, and automated operations—that's the second layer. In other words, the market traded the "replacement" narrative today but hasn't fully answered "who will be the first to productize AI capabilities and redefine where security budgets flow." These two are not the same thing. So what I'm more focused on is who can prove they are not a software company of the old era, but a security platform for the AI era. For leaders like $Palo Alto Networks(PANW.US), $CrowdStrike(CRWD.US), and $Zscaler(ZS.US), the greatest danger isn't short-term stock price pullbacks, but if customers shift more of their future budgets towards "native AI security capabilities," and these companies can only offer passive integration, then their valuation frameworks will continue to be compressed. Conversely, if these companies can turn AI from a feature point into part of their core architecture, this sell-off might not be a process of re-screening for winners. More crucially, Anthropic itself hasn't officially released Mythos; the full technical details, commercialization path, and final capability boundaries haven't been made public. The market selling off now shows capital already assumes the "capability leap" will impact the old logic first; but this kind of trading is also often the most prone to corrections once details materialize. In the end, what determines the medium-term direction of stock prices might not be whether Anthropic has a stronger model, but who can most quickly turn such models into security outcomes that customers are willing to pay for continuously. I won't just view today's red candle as emotional venting. I prefer to see it as a stress test that happened early: as AI moves from being an "assistive tool" to approaching a "fundamental capability layer," the valuation, product boundaries, and competitive dynamics of the cybersecurity industry will be re-examined. What got hammered today was stock prices; what will truly be tested next is the product definition power of these leading companies. Do you lean more towards this being a short-term emotional mis-kill, or is AI genuinely shaking the moats of traditional cybersecurity vendors? ### 相关股票 - [Okta (OKTA.US)](https://longbridge.com/zh-CN/quote/OKTA.US.md) - [Palo Alto Networks (PANW.US)](https://longbridge.com/zh-CN/quote/PANW.US.md) - [CrowdStrike (CRWD.US)](https://longbridge.com/zh-CN/quote/CRWD.US.md) - [Zscaler (ZS.US)](https://longbridge.com/zh-CN/quote/ZS.US.md) - [Datadog (DDOG.US)](https://longbridge.com/zh-CN/quote/DDOG.US.md) - [CloudFlare (NET.US)](https://longbridge.com/zh-CN/quote/NET.US.md)