The latest software market plunge comes after the announcement of Claude Code Security, capping off an unprecedented retrenchment across entire market sectors over a month and a half of after successive announcements by one company (and not the incumbent), through a coordinated, systematic expansion across the enterprise stack. Legal. Financial data. Knowledge work. IT services. DevSecOps. With each wave more targeted than the last, Anthropic went from “the serious alternative” to market mover.
No single company has ever moved markets quite like this. It was not a central bank decision, a geopolitical shock, or an earnings miss from a company that anchors an index. It was a sequence of product announcements from a private AI company, and the result was one of the most concentrated sectoral disruptions investors have ever seen, larger even than the initial launch of ChatGPT in late 2022.
By the time the third wave landed on February 20, the S&P 500 Software and Services Index had extended a losing streak to eight consecutive sessions and was down approximately 20% for the year. The WisdomTree Cloud Computing Fund fell roughly 20% in 2026. Salesforce and ServiceNow each lost about a quarter of their value.
The fact it is a private company doing the damage is a completely different facet that plays out not offensively but defensively. You can’t buy Anthropic stock, so the market responds with sell-offs instead. With a public company as the disruptor (Microsoft announcing Office 365, Amazon announcing AWS expansion) the market reaction is bilateral. Money rotates. You sell the threatened companies and buy the disruptor. The fear and the opportunity are in the same ecosystem. The selling pressure on incumbents is partially offset by buying pressure on the disruptor, and the overall panic is distributed.
Because Anthropic is private, there is no pressure valve. When Cowork plugins landed on February 3rd, every investor who read that announcement and believed it (who genuinely thought this changes the legal software market) had exactly one available action: sell Thomson Reuters, sell LegalZoom, sell RELX. They couldn’t buy the upside. The conviction had nowhere constructive to go. So it concentrated entirely into the sell side of the threatened categories.
This means the severity of the February selloffs was almost certainly amplified beyond what the fundamental threat justified, not because the threat wasn’t real, but because the market had no mechanism to rebalance. Normally disruption produces winners and losers in public markets simultaneously. Here it only produced visible losers. The winners were locked in a cap table accessible to a few hundred institutional investors and Anthropic’s existing backers.
There’s a second-order effect too. The analysts and fund managers watching this unfold couldn’t point to Anthropic’s revenue growth, user numbers, or market share gains as evidence the disruption was materializing. They had no stock chart to validate their thesis. So the fear remained speculative and narrative-driven longer than it would have with a public competitor, which paradoxically may have made it more volatile, not less, because it was never anchored to actual Anthropic fundamentals.
The sublimated buying demand likely also explains some of the exaggerated cascade — Intuit falling on a legal plugin announcement, advertising stocks falling on a marketing automation announcement. When the only move available is selling, investors apply the thesis broadly and quickly because there’s no portfolio construction discipline required. You’re not balancing a buy in Anthropic against a sell in ServiceNow. You’re just selling everything the narrative touches.
This is a structural argument for why the market reaction, while directionally right, was almost certainly oversized in magnitude, and why a correction or partial rebound was predictable once the initial narrative exhausted itself. The selling wasn’t fully rational price discovery. It was compressed sentiment with nowhere else to go.
It also raises an interesting question: when Anthropic eventually goes public, does that create a rebalancing moment? Does capital that’s been sitting on the sidelines, unable to express the “AI wins” thesis through Anthropic directly, rush in and partially restore valuations in the sectors it sold down? We don’t have a historical example to work from.
The Timeline: Three Waves in Six Weeks
January 12: Cowork Launches
We’ve looked at the seemingly overnight Anthropic surge within the AI industry, but the multisector impact on public markets bears its own scrutiny. Anthropic introduced Claude Cowork as a nontechnical counterpart to Claude Code, the same underlying architecture that had made Claude Code a dominant coding tool, now aimed at knowledge workers rather than software engineers. Available to premium subscribers at $100 per month, Cowork could execute multi-step professional tasks across documents, spreadsheets, presentations, and connected business software. The market’s initial reaction was muted. The tool’s implications had not yet been fully priced.
February 3–4: The Cowork Plugins and the First Selloff
The first significant market dislocation came on February 3 and 4, when Anthropic released industry-specific plugins for Cowork covering legal, finance, sales, marketing, data analysis, and customer support. The plugins were released under an open-source license. The market interpreted this as something more than a product update. It was the announcement that Anthropic was entering the core revenue verticals of established enterprise software companies directly.
The immediate casualties were concentrated in legal and financial data. Thomson Reuters fell 15.83%, its largest single-day drop on record. LegalZoom fell 19.68%. RELX, the British company that owns LexisNexis, fell 14%. London Stock Exchange Group fell nearly 13%. Gartner and S&P Global dropped 21% and 11% respectively. Intuit and Equifax each fell more than 10%. Advertising networks were hit on the news of marketing automation capabilities: WPP fell nearly 12%, Omnicom over 11%, Publicis 9%.
The contagion spread globally and immediately. In India, the Nifty IT index recorded its steepest intraday fall in years, with approximately 2 lakh crore in investor wealth erased in a single session. Infosys fell 8.23%, TCS fell 6.5%, Tech Mahindra fell 5.79%, Wipro fell nearly 4%. In Japan, NEC, Nomura Research, and Fujitsu each declined between 8% and 11%, dragging the Nikkei lower. In Australia, Xero fell sharply. The Nasdaq Composite declined 1.43%. The S&P 500 slipped 0.84%.
February 5: Opus 4.6 Compounds the Damage
Before the market had absorbed the Cowork plugin selloff, Anthropic announced Claude Opus 4.6, a model designed specifically to improve Cowork’s performance on knowledge work. Anthropic said it outperformed OpenAI’s GPT-5.2 on benchmarks evaluating AI performance in finance and legal work, directly the sectors that had just been hit hardest. The model featured an expanded context window of one million tokens, coordinated multi-agent teams mimicking human engineering team structures, and a new PowerPoint integration capable of generating production-ready slides. Anthropic said documents, spreadsheets, and slides produced by the model would be “closer to production-ready on the first try, meaning they should require less human intervention.”
The financial data sector bore the brunt of the second wave. FactSet fell 10%. S&P Global, Moody’s, and Nasdaq all saw renewed sharp declines. The Nasdaq just had its worst two-day tumble since April. ServiceNow and Salesforce each dropped nearly 7% that session. The software ETF had its worst two-day stretch since April.
February 20: Claude Code Capabilities and Claude Code Security
On February 20, Anthropic made two announcements simultaneously. The Claude Code desktop app received significant new workflow capabilities: it can now launch development servers, display running applications directly in the interface, read console logs, monitor pull requests in the background, automatically attempt to fix CI failures, and merge pull requests once checks pass. A developer can open a PR, move to an entirely different task, and return to find the first issue resolved. Sessions now sync across devices and platforms.
The second announcement was more pointed. Anthropic unveiled Claude Code Security, a purpose-built AI vulnerability scanner now in limited research preview for Enterprise and Team customers. This was not a feature update. It was a direct competitive entry into the application security market. Anthropic claimed the tool had discovered over 500 security flaws in production open-source codebases, including vulnerabilities that had gone undetected for decades. The tool runs findings through multi-stage verification where the model argues against its own conclusions to filter false positives, assigns severity ratings, and suggests patches, all with developer approval required before anything ships. The company stressed that offensive and defensive capabilities advance in parallel, and that getting Claude Code Security into defenders’ hands first was a deliberate choice.
CrowdStrike fell 8%, Cloudflare fell 8.1%, Zscaler dropped 5.5%, SailPoint shed 9.4%, Okta fell 9.2%. The Global X Cybersecurity ETF fell 4.9%, closing at its lowest level since November 2023.
In three waves across six weeks, Anthropic’s announcements had hit legal software, financial data providers, enterprise SaaS, advertising conglomerates, global IT services across three continents, professional services platforms, and the entire cybersecurity sector. What started as concern about knowledge worker automation had expanded into a fundamental reassessment of where the software value chain survives in an AI-driven enterprise environment.
Why Each Wave Hit What It Hit
The precision of the sectoral damage is worth examining, because it was not random. Each announcement landed where it did because of what it actually threatened.
The Cowork plugins threatened the revenue models of companies whose core product is structured information retrieval and professional task execution. Thomson Reuters and LegalZoom are, at their core, selling access to a capability (legal research, contract assembly, compliance review) that an AI agent with deep context and legal reasoning can now perform within a general subscription. The same logic applies to LexisNexis, Wolters Kluwer, FactSet, and the financial data vendors. Their moats were built on the cost and difficulty of assembling and accessing specialized information. Claude does not need to access their databases. It has internalized sufficient domain knowledge to perform the same workflows.
Advertising held a different logic. Anthropic’s marketing automation plugin threatens not just software tools but the billable workflow of agencies themselves (campaign planning, copy generation, audience segmentation, performance reporting). The Publicis and WPP declines reflected that the marketing automation vertical these companies had been investing in was being directly commoditized.
The Indian IT selloff is structurally distinct. Infosys, TCS, and Wipro are headcount businesses. They bill clients for the hours of software engineers, business analysts, and process workers executing tasks. Application services (coding, testing, documentation, data processing, analytics) constitute 40% to 70% of revenues for many of these firms. Claude Code automates coding. Cowork automates business process work. As Jefferies noted in a contemporaneous research note, the revenue exposure is direct and the threat to utilization rates is real. The Nifty IT index fell approximately 9.4% over the week ending February 7, its worst weekly performance in over 10 months. By mid-February, a subsequent selloff erased another $12 billion in sector market cap in a single session.
The cybersecurity reaction on February 20 was the most precisely targeted. Unlike the earlier waves, which reflected fear of general capability expansion into adjacent verticals, the Claude Code Security announcement was a direct competitive product launch: a named tool, with named capabilities, available to named customer tiers, going head to head with the core function of application security vendors. The market did not need to extrapolate. It was reading a product announcement.
The Analytical Case: What These Announcements Actually Signal
The surface-level interpretation (AI replaces software) is both directionally correct and analytically incomplete.
Cowork and its plugins do not replace the entire enterprise software stack. As Gartner noted in a February research note, they are “potential disrupters for task-level knowledge work but are not a replacement for SaaS applications managing critical business operations.” Wedbush Securities made the same point more bluntly: enterprises will not overhaul decades of software infrastructure investment to migrate to Anthropic’s platform. The switching costs, compliance requirements, and institutional inertia are real and significant. But as we have argued previously, it changes where the value sits.
What the announcements do threaten is more targeted but still consequential: the pricing power and growth rates of companies whose value proposition rests on task-level execution in bounded professional domains. If Claude can review a contract, generate a compliance summary, and produce a financial model with less human intervention than the software that currently does those things, the enterprise does not need to fully migrate, it just needs to stop expanding its seat count. And seat count expansion has been the primary growth engine for most of the companies the market punished.
The cybersecurity dimension is distinct in another way. Claude Code Security does not just compete with vulnerability scanning tools. It introduces a new architectural reality: an AI agent operating at machine speed inside the development loop, with reasoning capabilities that go beyond pattern matching to semantic analysis of how code components interact. The same system that can find vulnerabilities can theoretically be prompted to exploit them. Anthropic acknowledged this directly: the company’s Frontier Red Team has been testing offensive capabilities through capture-the-flag events and collaboration with national laboratories precisely because the dual-use risk is real and present. The release of Claude Code Security is a bet that defenders moving quickly creates net societal benefit. It is also an admission that the arms race has already started.
For the broader software market, the more important signal is architectural. Anthropic has now established Claude as a platform that operates across the full enterprise technology stack — from raw code generation and CI pipeline management at the developer layer, to knowledge work automation at the business user layer, to vulnerability detection at the security layer. That is not a point solution. It is a horizontal capability platform that can embed itself at every layer where enterprise software currently extracts rent.
Historical Precedents: Has This Happened Before
The February 2026 selloff belongs to a specific category of market event that recurs at inflection points in technology infrastructure. The mechanism is consistent: a platform player with significant distribution and technical capability announces entry into adjacent markets that had been comfortably served by standalone specialists. Markets price the risk immediately and often excessively in the short term, while the actual displacement plays out over years.
The iPhone, 2007. When Steve Jobs announced the iPhone in January 2007, Nokia and Research In Motion fell sharply within sessions. The damage looked immediate but was actually slow. Both companies continued selling devices in large volumes for several years. Nokia’s stock peaked near €26 in 2007 and fell below €10 within two years; its handset division was eventually sold to Microsoft for a fraction of its former value. By 2013, Apple’s market cap had reached roughly $3 trillion while Nokia’s had fallen to $16 billion from approximate parity in 2007. The iPhone did not just disrupt handset makers. Camera makers, GPS device manufacturers, portable music players, mobile game hardware, and eventually carriers themselves all felt the reorganization. A single platform absorbed the functions of dozens of adjacent categories. The key structural parallel is this: the iPhone’s announcement effect and its fundamental effect operated on different timescales. The market was directionally right in January 2007. It was early on magnitude and timing.
AWS, 2006. When Amazon launched S3 in March 2006 and EC2 in August of the same year, the market did not react at all. A Piper Jaffray analyst said the investments were “probably more of a distraction than anything else.” The companies most exposed (enterprise hardware vendors, managed hosting companies, large IT services firms) did not price in the disruption at announcement. They priced it in slowly, across years, as AWS achieved what Jeff Bezos later described as “the greatest piece of business luck in the history of business” — seven years of building without like-minded competition, largely because incumbents did not believe a bookseller could be a credible enterprise software company. The AWS case is the counterexample to the iPhone pattern: when the disruption requires technical sophistication to assess, the market is slow. The February 2026 reaction suggests that this time, the market is not making the AWS-era mistake. Financial analysts covering software companies in 2026 understand AI capability claims better than analysts covering server hardware understood cloud computing in 2006. The immediate and global pricing of the Anthropic announcements reflects that comprehension gap closing.
ChatGPT and Chegg, 2022–2023. This is the most structurally precise historical parallel to the Cowork and plugins dynamic. Chegg’s business monetized access to a large proprietary database of solved academic problems. It had 6.2 million subscribers paying up to $19.95 per month, with an estimated 36% penetration rate among U.S. college students, an extraordinary moat by any metric. ChatGPT did not announce that it was competing with Chegg. It launched as a general-purpose AI assistant. For months after ChatGPT’s November 2022 launch, Chegg’s stock held relatively stable. The reckoning came in May 2023 when Chegg’s CEO disclosed on an earnings call that student interest in ChatGPT had spiked in March and was now materially affecting new customer growth. Chegg fell 50% in a single session. The stock ultimately lost 99% of its peak market capitalization, erasing $14.5 billion in value. The same day Chegg disclosed its ChatGPT impact, LegalZoom fell 8% and Duolingo fell 10% on no news of their own; the market had identified the category risk and applied it broadly. That cascade logic is precisely the pattern visible across the six-week Anthropic sequence: each announcement triggered damage in the directly threatened sector and collateral damage in adjacent ones. The Chegg case also illustrates the key variable determining survival. Chegg’s moat was the cost and difficulty of accessing a database of answers. Once a foundation model could answer the same questions for free, the moat did not erode, it evaporated. Companies whose moat is genuinely proprietary data, deep client relationships, or governance infrastructure have more durable positions.
Microsoft Office bundling, 1993–1995. Lotus 1-2-3 and WordPerfect were market leaders in their respective categories (spreadsheets and word processing) with large installed bases and premium recurring revenue from corporate clients. When Microsoft began bundling Word, Excel, and PowerPoint into a single suite priced below the cost of individual components, the standalone market for productivity software collapsed over roughly three to four years. Lotus and Novell (which acquired WordPerfect) were both eventually sold or wound down. The disruption mechanism was not superior product quality. Microsoft’s applications were competitive but not obviously superior at the time of bundling. It was distribution and pricing leverage from the platform. The parallel to Anthropic’s approach is not exact, but structurally similar. By including legal, financial, and marketing automation plugins within the existing Claude subscription, Anthropic is pricing general-platform capability against the standalone price of specialized tools. Enterprises do not need to fully substitute, they need only decide that incremental expansion of the standalone product is no longer justified. When the value lives in the data, value that can now be extracted most efficiently by the exact kind of coordinated, comprehensive corporate AI Anthropic now exemplifies, the software casing is secondary.
What the Precedents Suggest About the Outcome
Taken together, these cases point to a consistent pattern in how platform disruption moves through markets and then through industries. But it hasn’t ever happened that so many industries have been impacted by one company in such a short time frame.
Markets react immediately and directionally. The magnitude of the initial reaction is often excessive — the sell-off in application security stocks on February 20 is probably pricing in a pace of displacement that will take years to materialize. Wedbush Securities, Jensen Huang, and Gartner are all essentially correct that the “SaaSpocalypse” framing is an overstatement of short-term impact.
But the directional signal is almost certainly right. In each of the historical cases, the initial market reaction captured a real architectural shift even when the immediate revenue impact was minimal. Nokia did not lose material smartphone market share in 2007. Chegg did not lose significant subscribers in November 2022. AWS did not take meaningful share from enterprise hardware vendors in 2006. The displacement in all three cases came later, through the progressive realization that the cost and quality gap between the platform and the specialist was widening rather than narrowing.
The companies most at risk across the categories Anthropic has entered share the same characteristic that made Chegg, WordPerfect, and the smartphone incumbents vulnerable: their moat was primarily the cost and difficulty of accessing a function, not the depth of proprietary insight or relationship advantage underlying it. Vulnerability scanning tools that rely on pattern matching, legal software that assembles standardized documents from templates, IT services firms billing for headcount on well-defined code maintenance tasks — all of these are exposed for the same structural reason.
The companies with more durable positions are those whose value is genuinely irreplaceable by the general-intelligence layer: proprietary data that Claude has not ingested, deep compliance and regulatory expertise, audit trails and governance infrastructure, incident response capabilities requiring human judgment under adversarial conditions, and — increasingly — the governance systems required to manage AI agents themselves. As Claude Code Security and Cowork expand the surface area of autonomous AI action inside enterprises, the need for oversight, traceability, and containment does not decrease. It increases. That is likely where the next generation of software value concentrates.
The February 2026 market reaction is, in historical terms, the moment investors began pricing the transition. Like every such moment in technology cycles, it is happening faster than the fundamental displacement and slower than the underlying shift in architecture. The questions now are how the ecosystem rebalances asit always does, whether the incumbents in each category adapt faster than the platform expands, and what impact a public Anthropic will eventually have.
