Cybersecurity stocks fall after Claude Code Security presentation

Is artificial intelligence devouring the tech industry from the inside? The discreet launch of a tool by Anthropic was enough to cause billions of dollars in stock market losses. And this may just be the beginning.

When AI attacks cybersecurity

On February 20, 2025, Anthropic deployed in preview Claude Code Security, a code analysis tool based on its flagship model Claude Opus 4.6.

The promise is clear: scan the entire codebase, identify vulnerabilities, validate each result to limit false positives, then suggest fixes. All “like an experienced security researcher,” according to the company’s own words.

The initial results are striking. Claude Opus 4.6 has reportedly already detected over 500 high-severity vulnerabilities that had withstood decades of human auditing, according to VentureBeat. OpenAI also crowned it the best model on its benchmark dedicated to smart contract security, launched on February 19.

The markets reacted harshly. CrowdStrike lost 18% in a few days, wiping out 20 billion dollars. Palo Alto Networks, a sector giant with 116 billion in market capitalization, fell 9%. Fortinet, Cloudflare, and Zscaler followed the same path. In one week, the entire publicly traded cybersecurity sector came under pressure.

According to Shrenik Kothari, analyst at Robert W. Baird, this is “a panic-driven massive sell-off and a dominant narrative.” But other voices nuance this interpretation.

An earthquake beyond cybersecurity

The Kobeissi Letter stated it bluntly in an analysis published Tuesday: “When AI reproduces employees’ work, pricing power shifts to the buyer.” A short sentence but that sums up the fundamental issue of this disruption.

Because the shockwave did not stop at cybersecurity. IBM suffered Monday its biggest daily drop since October 18, 2000, with a 13.2% decline. 

The reason? Anthropic published a blog post explaining that Claude Code can automate the modernization of COBOL, the old programming language that still runs banking, insurance, and government systems on IBM mainframes. Where armies of consultants took years, AI promises to reduce the project to a few quarters.

Wedbush analysts speak of “fears of AI-related phantom transactions” but paradoxically acknowledge that Anthropic strengthens the idea that cybersecurity will be one of the main beneficiaries of AI’s rise, no longer an external service provider but integrated directly into development tools.

What this sequence reveals is less a passing crisis than a structural realignment. AI no longer just competes with humans; it now attacks software itself. For investors as well as tech companies, the question is no longer whether AI will reshuffle the cards, but how fast.