During the Chinese New Year, the AI circle has been experiencing successive earthquakes. Yesterday's Gemini 3.1 Pro was just unveiled, and then came another bomb - just now, Anthropic released a code security tool, Claude Code Security, which can efficiently scan code base vulnerabilities and automatically generate targeted patches, far exceeding traditional tools.
Does it sound like just a mediocre technical update?
As a result, as soon as the news came out, the cybersecurity sector of the U.S. stock market immediately plunged collectively.
Security stocks such as CrowdStrike, Cloudflare, and Okta plummeted by more than 5% in an instant, and their total market value evaporated by more than 10 billion U.S. dollars. Tens of billions of market values were evaporated overnight!
Take a closer look, the leading network security company CrowdStrike once plummeted 6.5%, Cloudflare fell more than 6%, SailPoint fell 6.8%, Okta fell 5.7%, Zscaler fell 3.5%, Global X Cybersecurity ETF fell 3.8%, and the cumulative decline during the year expanded to 14%.
An AI function update directly caused a 100-billion-dollar industry sector to bleed out on the spot.
This is not volatility, this is panic!
Foreign media quickly published: The stock price of the network security company plummeted on Friday simply because Anthropic introduced a security feature in the new Claude model.
Security stocks are booming across the board, and the market value of giants has shrunk by tens of billions
As soon as Anthropic's new killer tool was released, the network security market tasted blood.
The share price of CrowdStrike, the leading network security company, fell by more than 6.5%. Cloudflare and Okta followed suit, and the total market value evaporated by more than 10 billion US dollars overnight.
Global investors are in a panic: AI will directly eat into the market share of professional security manufacturers!
The Cybersecurity ETF fell 4.9%, closing at its lowest level since November 2023.
In the first few weeks of 2026, market anxiety continues to rise.
The iShares Expanded Technology Software ETF has fallen more than 23% this year and is on track to post its largest quarterly decline since the 2008 financial crisis!
And ironically, Cloudflare was once seen as a beneficiary of the rise of Anthropic; at the end of January this year, its stock price rose sharply after the adoption rate of an open source AI assistant based on the Claude model increased.
However, the market is changing rapidly.
You know, half of the startups in YC are working in the security direction. Once the new tool Anthropic is released, will they still survive?
Whenever Claude releases a new version, SaaS business owners feel like this:
A Zheng I directly overthrew us with the overwhelming force!
Claude transforms into the most powerful white hat hacker
Why did Claude Code Security cause the collapse of global security stocks?
The reason is that it is not only a tool, but also a complete subversion of the traditional security defense system.
While the security team was still worrying about the mountain of vulnerability tickets, Claude had transformed into the "strongest white hat hacker" and jumped into the open source code base with the just-released Claude Opus 4.6.
The results were jaw-dropping: it uncovered more than 500 epic bugs that had been dormant for more than ten years!
You must know that these vulnerabilities have been reviewed line by line by countless top human experts, and they are still "slip through the net" in the code.
Dimensionality reduction strikes, traditional scanning tools overturned?
The pain points in the security community have always been extremely painful: loopholes can never be fixed, and there are never enough security experts.
Everyone is using traditional static analysis tools (SAST), but these tools are often based on rigid rule matching.
It's okay to look for hard-coded passwords, but once you encounter complex business logic flaws or unauthorized vulnerabilities, you'll be blind.
The emergence of Claude Code Security has completely changed the rules of the game.
It does not engage in rigid pattern matching, but rather like a human security expert with ten years of experience, it truly "reads" and "deduces" your code!
It can deeply understand how various components love and kill each other, and follow the flow of data in the application.
Complex vulnerabilities that cannot be seen by rule scanning tools have no way of hiding in front of its cheat-like logical reasoning.
Never change it blindly! Self-validation against hallucinations
Everyone is afraid of hallucinations when writing AI code, but what about AI checking loopholes and fixing bugs?
Anthropic is incredibly stable this time. In order to filter out troublesome "false positives" (false positives), every discovered vulnerability must undergo an extremely stringent "internal multi-stage verification."
Claude will force himself to play both red and blue, desperately trying to prove or disprove his findings.
The truly high-risk vulnerabilities that finally stand out will be pushed to the security dashboard. Not only does it automatically generate repair patches with ratings, it also comes with a detailed "confidence index."
The most important thing is that Claude adheres to the restraint of "only giving advice, not doing anything for you". Identify the problem, provide the solution, but it is always the human developer who ultimately presses the merge button.
Excellent performance, personally tested by the red team
The power of this system is not achieved overnight.
Over the past year or so, Anthropic's Frontier Red Team has literally pushed Claude to the ground for crazy, high-intensity training.
Not only did it enter the CTF security competition to compete with human hackers, it even cooperated with the Pacific Northwest National Laboratory to use AI to defend national critical infrastructure.
Real knowledge comes from actual combat. It is this kind of "hell-level" special training that gives Claude Opus 4.6, released earlier this month, such terrifying defense power.
Those more than 500 hidden loopholes that were uprooted are the best military medals.
Even Anthropic himself admits: “We also use Claude to review our own code on a daily basis, and the effect is amazing!”
AI Offensive and Defense War: Nuclear Deterrence Level Arms Race Begins
There is no doubt that this is a turning point of great historical significance.
In the near future, most of the world's code will be scanned over and over again by AI.
Hackers will surely use AI crazily to mine exploitable weaknesses on a large scale. But as the saying goes, a defensive team that moves faster can use the same or even stronger AI magic to nip loopholes in the cradle.
Currently, Claude Code Security has opened a limited research preview version to Enterprise and Team customers!
The maintainers of open source projects also enjoy the expedited channel of “green light all the way”.
Whoever can master the trump card of AI security first will be invincible in this life-and-death competition between computing power and intelligence.
AI eats up the middle layer, what the capital market fears most is coming!
In the past two years, AI coding has made programmers nervous. But everyone is still comforting themselves: AI can only write code and fix bugs. In highly specialized fields such as security, it cannot replace humans.
As a result, now, Claude directly started scanning for vulnerabilities.
This means that AI no longer generates content, but enters the core workflow of enterprise security. It directly targets the sky-high profit pool of enterprise-level security services!
What is the capital market most afraid of? It’s not the technology, but the pricing power that has been shaken.
The reason why network security companies have long-term high valuations is because offensive and defensive confrontations are complex and security experts are scarce, which makes their services highly specialized.
As a result, now that Claude Code Security is released, an AI model can complete 80% of vulnerability scanning and repair recommendations, and enterprises only need a small number of security engineers.
So, do enterprises still need to pay high subscription fees to security vendors?
This is why Claude has already caused a sell-off in the market even before it can replace CrowdStrike.
When investors asked, "Will we still need so many security companies in five years?", the stock market began to collapse!
It seems like just yesterday that Anthropic used Claude to trigger a trillion-dollar sell-off
Looking back at the series of nuclear explosion-level incidents triggered by Claude Code at the end of last year, we can find that the most cruel thing about this type of AI revolution is that it first devours the middle layer.
Once the model accuracy is high enough, SaaS premiums will be compressed, service fees will be revalued, and valuation logic will be broken!
What’s even more frightening is that don’t forget, Anthropic said that this feature is only a “limited research preview” and has not yet been fully opened and has not been verified on a commercial scale.
But the stock price has plummeted, which shows that the current evolution speed of AI is much faster than the product iteration speed of traditional software companies.
This is a dangerous signal - something more terrifying may happen next.