There is a particular kind of fear that spreads through financial markets faster than any other. Not the slow dread of a recession building over quarters, not the gradual erosion of a business model over years. It is the sudden, gut-punch realization that something you thought was safe is no longer safe at all. That is what Anthropic handed Wall Street twice in the span of three weeks in February 2026, and neither time did the market take it quietly.
This is the story of a relatively small AI company(Which aint small anymore) that does not even trade publicly, and how it has managed to shake billions of dollars out of some of the most established technology firms on earth, simply by releasing new software.
The Day Cybersecurity Stocks Forgot How to Stand Up
On February 20, 2026, Anthropic announced Claude Code Security. The name sounds tame enough. It is a tool built into the company’s Claude Code platform that scans software codebases for vulnerabilities and proposes targeted fixes. No human needs to approve every line it reads, but every fix it suggests still requires a developer’s sign-off before anything changes. Responsible, measured, cautious.
Markets did not care about the caution.
CrowdStrike fell 8 percent. Cloudflare lost 8.1 percent. Okta dropped 9.2 percent. SailPoint shed 9.4 percent. TheGlobal X Cybersecurity ETF closed at its lowest level since November 2023. JFrog had one of the worst days among the group, losing 24 percent, while GitLab fell more than 8 percent and Palo Alto Networks saw notable declines as investors reacted to potential disruption in the 2.5 billion dollar AI coding market.
What made investors panic was not just what the tool did. It was what it had already done. Anthropic said Claude Opus 4.6, the model powering the tool, had found over 500 vulnerabilities in live open-source codebases during internal testing, some of which had been sitting there for decades without any expert catching them. Read that again slowly. Decades. These were not toy projects. These were real, production-grade systems that had survived years of professional scrutiny. Claude found what human reviewers had missed, and it found a lot of them.
For investors holding shares in companies whose entire business is built on finding and neutralizing exactly those kinds of threats, that sentence was a very uncomfortable read.
What Claude Code Security Actually Does Differently
To understand why the market reacted so sharply, you need to understand the fundamental difference between what existing security tools do and what Claude Code Security claims to do.
Traditional static analysis tools compare code against a library of known vulnerability patterns. This catches obvious problems like exposed credentials or outdated encryption, but it is unreliable for subtler issues like business logic errors or faulty access controls. Think of it like a spell-checker that only flags words it recognizes as wrong. It is fast and useful, but it cannot understand intent.
Anthropic’s tool is designed to read and reason about code the way a human security researcher would. It understands how components interact, tracks how data flows through an application, and spots the complex vulnerabilities that rule-based tools miss. Each finding goes through a multi-stage verification process before it reaches an analyst, with Claude revisiting its own conclusions to filter out false positives. The results appear in a dashboard where engineers can review findings, inspect proposed patches, and approve fixes. Nothing gets applied automatically.
The company built this over more than a year of deliberate research, running its models through capture-the-flag competitions and a partnership with the Pacific Northwest National Laboratory to test defenses for critical infrastructure. This was not a rushed feature announcement. It was a calculated product with serious foundations.
The Irony Nobody Could Ignore
Here is where the story gets genuinely uncomfortable.
The same Claude Opus 4.6 model now positioned as a security defender was blamed just days earlier for a 1.78 million dollar loss at DeFi lending protocol Moonwell. The model that found 500 bugs humans missed had also, in a different context, been implicated in an exploit that caused real financial damage.
Anthropic had actually anticipated this duality. The company’s internal research, published in December 2025, showed that an earlier version of the model could independently identify and exploit smart contract vulnerabilities worth up to 4.6 million dollars in a controlled setting, with minimal human involvement. Anthropic knew its models could cut both ways. Claude Code Security was its attempt to get defenders armed before attackers figured out the same tricks.
Anthropic acknowledged the trend directly, warning that “less experienced and resourced groups can now potentially perform large-scale attacks,” and stating that “attackers will use AI to find exploitable weaknesses faster than ever.” The argument is essentially: this capability exists regardless of what we do, so let us put it in the hands of defenders first.
Whether you find that argument reassuring or troubling probably depends on your risk tolerance.
This Was Not the First Time in February
The cybersecurity selloff was actually the second major market shock Anthropic caused that month. The first happened on February 3, 2026, and it was considerably larger.
A new AI automation tool from Anthropic sparked a 285 billion dollar rout across software, financial services, and asset management sectors as investors raced to dump shares with even the slightest exposure. A Goldman Sachs basket of US software stocks sank 6 percent, its biggest one-day decline since April’s tariff-fueled selloff.
The trigger was Claude Cowork, a nontechnical version of Claude Code designed for office workers rather than programmers. Anthropic had released industry-specific plugins that let users tailor the tool for legal, finance, sales, and marketing tasks, sparking fears that the technology could replace specialized research and financial analysis software.
Thomson Reuters fell by 15.83 percent on that Tuesday. LegalZoom fell by nearly 20 percent the same day. The legal software sector, an industry that had spent years building moats around specialized knowledge and established workflows, suddenly found itself staring at a general-purpose AI that could, in theory, do a large portion of what its products do.
Then, on February 5th, Anthropic released Claude Opus 4.6, and markets slipped further.
What Opus 4.6 Changed and Why Software Companies Started Sweating
Claude Opus 4.6 was not just an incremental model update. It expanded the context window from 200,000 tokens to one million, allowing Claude to process and reason across vastly more information at once. In practical terms, a developer could now feed Claude an entire enterprise codebase and the model would hold it all in mind simultaneously, tracking patterns and relationships across hundreds of thousands of lines.
But the feature that sent the most anxiety through the software industry was Agent Teams. This feature lets autonomous teams of AI agents tackle complex projects together, allowing users to deploy multiple agents simultaneously that handle different aspects of a larger project, working in parallel and communicating with each other to coordinate their efforts. It mimics how human teams divide up work across a big assignment.
For companies like Salesforce, Workday, and SAP, this was not an abstract threat. Their entire business model rests on human users paying subscription fees to access their tools. If AI agents can interact with software by navigating interfaces themselves, the logic of per-seat licensing begins to look fragile. Salesforce shares shed nearly 30 percent year-to-date as investors feared what analysts began calling “seat churn.”
Intuit declined 32 percent while Thomson Reuters fell 30 percent over the same period. The WisdomTree Cloud Computing Fund lost more than 20 percent year-to-date. These are not small moves.
The Bull Case: Why Some Analysts Think the Fear Is Overblown
Not everyone who watched these selloffs thought the market was being rational.
Wedbush analyst Dan Ives, who covers technology stocks closely, pointed out that large organizations have ingrained workflows and processes that cannot simply be switched overnight to new AI tools. The friction of enterprise software adoption is not just technical. It is organizational, contractual, regulatory, and cultural.
Analysts also noted that Anthropic’s new capability is relatively narrow. The tool scans code for vulnerabilities and suggests fixes for human review. It is designed to help development teams identify weaknesses that conventional tools might miss. It is useful, but it is not a replacement for cybersecurity platforms.
CrowdStrike, for example, does cloud-based endpoint protection, antivirus capabilities, and real-time threat response to active attacks. Okta and SailPoint manage identity and user behavior across complex enterprise environments. Claude Code Security works at the development stage, before software is ever deployed. These are genuinely different problems.
The data from real AI deployments is also more sobering than the market panic suggests. MIT research found that 95 percent of AI pilots failed to deliver meaningful results, and BCG reported that only 5 percent of companies that deployed AI saw measurable value from it. The gap between what a model can do in a demo and what it reliably delivers in a messy enterprise environment remains wide.
Nvidia CEO Jensen Huang argued that older software companies possess protective advantages, including specialized products, massive data repositories, and existing AI adoption. Companies like Salesforce are not sitting still. They are pivoting aggressively toward AI agent platforms, experimenting with outcome-based pricing, and building their own AI layers into established products.
The Bear Case: Why Some Analysts Think the Fear Is Exactly Right
That said, dismissing investor anxiety as a knee-jerk overreaction misses something real.
The traditional SaaS business model assumes that software is complicated enough that companies need to pay experts to build and maintain specialized tools for specific functions. Legal research software exists because legal research is hard. Financial analysis platforms exist because financial analysis requires specialized tools. Security scanners exist because finding code vulnerabilities requires deep expertise.
AI systems are attacking every one of those justifications simultaneously. The sentiment shift was palpable as Anthropic’s new models proved they could interact with software via the backend or by visually navigating the UI, making the user experience of many legacy platforms secondary to the API’s efficiency. When an AI can navigate your software’s interface on behalf of the user, the value of having a beautiful, intuitive interface drops considerably.
Enterprise large language model spending reached 7 million dollars on average per company in 2025, a 180 percent jump from 2.5 million in 2024, with projections of 11.6 million for 2026. The money is real and accelerating. Every dollar a company spends on AI tooling is a dollar that might not go toward renewing a traditional SaaS subscription.
The long-term question is not whether AI will eventually replace significant parts of what software companies do. Most thoughtful observers believe it will. The question is the pace, and whether incumbents can adapt fast enough to retain their customers through the transition.
Future Expectations: What Comes Next for the Industry
The next twelve months will test whether the market’s pricing of this disruption is ahead of reality or behind it.
Anthropic itself will be a key variable. The company is currently valued at 350 billion dollars despite being private, and reports suggest it is exploring an IPO. Claude Code already reached a 1 billion dollar run rate in revenue six months after launch. The speed of adoption suggests that what happened in software coding is likely to repeat in other professional domains.
The cybersecurity sector faces a particularly interesting inflection point. The market sold off hard at the announcement of a tool that is currently only available in limited research preview to enterprise and team customers. If Claude Code Security moves into broad availability and the 500-vulnerability finding rate holds up in production deployments, the competitive pressure on traditional security vendors will become far more concrete.
For traditional software companies, the adaptation playbook is becoming clearer even if execution remains difficult. Companies that own proprietary data, that have deeply integrated their platforms into enterprise workflows, and that can credibly offer AI-native versions of their tools have a real chance of surviving. Companies whose value proposition rests purely on the complexity of the software itself, rather than on the data or relationships underneath it, face a genuinely difficult road.
Regulatory scrutiny will also matter. The same capability that lets Claude find decades-old bugs also raises uncomfortable questions about what happens when AI-generated security patches introduce new vulnerabilities, or when autonomous agents begin touching sensitive enterprise systems. Governments in the US and Europe are watching, and the rules are still being written.
The Competition: Where OpenAI, Gemini, and Others Stand Right Now
It would be a mistake to read all of this as an Anthropic story in isolation. The broader AI race is reshaping the competitive landscape for every technology company, and the participants are running hard.
OpenAI remains the largest player by revenue. The company is on a 20 billion dollar annualized revenue run rate as of late 2025, with Anthropic following at around 4 billion dollars in ARR but projecting growth toward 18 billion by the end of 2026. OpenAI released its own coding assistant on the same day as Opus 4.6, intensifying what has become a weekly release cycle across the industry.
On the enterprise front, Anthropic holds roughly a third of the enterprise AI market according to survey data, compared with 25 percent for OpenAI and about 20 percent for Google Gemini. That is a remarkable position for a company that barely existed in enterprise software discussions two years ago.
Google is playing a different game. Gemini has surged from 5.4 percent of the AI chatbot market share in January 2025 to 18.2 percent by early 2026, while ChatGPT’s share declined from 87.2 percent to 68 percent over the same period. Google’s advantage is not purely technical. It is distributional. Gemini sits inside Android, inside Gmail, inside Google Docs. It does not have to convince users to show up. The users are already there.
The battle in 2026 is no longer purely about who can build the better model. It is about distribution, monetization, and cost efficiency. Google’s ability to embed AI into products that already command billions of users gives it a structural edge that pure AI companies cannot easily replicate.
OpenAI, meanwhile, faces pressure from multiple directions. Sam Altman’s company is valued at 500 billion dollars and has made more than 1.4 trillion dollars in infrastructure commitments, a staggering sum that has raised serious questions about how the company will be able to afford those obligations. The financial engineering required to make that math work is considerable.
What the competitive picture reveals is that no single company is going to “win” AI in the way that Google won search or Microsoft won enterprise productivity. Anthropic and Google are building the agentic protocols that form the scaffolding of the future of agentic AI, with the Model Context Protocol having become something like a universal standard for AI integrations. The infrastructure of AI is being built collaboratively and competitively at the same time, which is a strange thing to witness.
The Fear Is Real, But So Is the Nuance
Something genuinely significant is happening in the software industry. That much is not debatable. What is debatable is whether every stock that sold off deserves the valuation it now carries, whether every threat priced in will actually materialize, and whether the companies being disrupted have more resilience than panicking investors are willing to credit.
What Anthropic demonstrated in February 2026 is that a product release from a private AI company can now move public markets in ways previously reserved for central bank decisions and major geopolitical events. That is new. That deserves attention.
The cybersecurity companies that sold off hard have not stopped doing what they do. The legal software firms that lost a fifth of their value in a single day still have customers, contracts, and teams. The financial analysis platforms still have proprietary data that Claude cannot access from a public release.
But the window of time they have to adapt is shrinking, and the pressure to prove that AI makes them better rather than obsolete is becoming the central question of enterprise software.
Anthropic did not create that pressure. It just made the timeline feel very, very immediate.