The AI industry’s central question has shifted from “how fast can we push performance higher” to “how do we safely control rapidly evolving superintelligence and AI agents,” leading cybersecurity to be revalued as a critical new growth engine of the AI era. In a striking paradox, a security sector that was grappling with growth stagnation just months ago has emerged as the biggest beneficiary of the AI slowdown debate.
On the 14th (local time), CrowdStrike (CRWD) shares surged approximately 15% to a record high on U.S. markets, while Palo Alto Networks (PANW) climbed roughly 14%. Other cybersecurity companies including Zscaler also posted double-digit gains. In contrast, AI semiconductor stocks such as Nvidia (NVDA) and AMD fell as concerns over an AI development slowdown spread.
The trend became even more pronounced after Jensen Huang, CEO of Nvidia, predicted at the Goldman Sachs Communacopia Technology Conference on the 10th that cybersecurity would become the next major growth market for AI. According to Seeking Alpha, he directly referenced cybersecurity’s growth potential to investors.
Attack speeds accelerate while defense stagnates
The biggest reason the AI slowdown debate presents an opportunity for security companies is the widening gap between development speed and security verification speed. According to the latest report from U.S. security firm Veracode, source code automatically generated by AI coding tools passes security testing at a rate of just 56%. While AI model performance itself has improved dramatically, the code vulnerability pass rate has remained stagnant for months.
In response, Anthropic’s Claude Code and OpenAI’s Codex are racing to build real-time integration systems that embed security verification directly into the development environment. The speed at which attackers are adopting AI is equally alarming. According to CrowdStrike, cyberattacks leveraging AI have surged 89% year-over-year. The average “Breakout Time” — the time it takes a hacker to expand access from an initial breach to other internal systems — has shrunk 65% year-over-year to 29 minutes, with the fastest recorded time being just 27 seconds.
The proliferation of AI agents that autonomously perform enterprise tasks is fundamentally reshaping the security landscape. As AI agents with email-reading capabilities, internal database access, and code execution privileges multiply, the security perimeter that once centered on servers and networks is now expanding to encompass AI prompts, models, agent accounts, and external tools.
Anthropic disclosed to Reuters on the 14th that it had detected and blocked a Russia-linked hacking group that exploited its Claude model to attack more than 20 Ukrainian government and defense-related targets. The attacks, which ran from December of last year through August of this year, employed phishing, hotel Wi-Fi network compromise, and WhatsApp account breaches. Anthropic noted that AI performed a substantial portion of the work while humans primarily served in a supervisory role.
According to joint research by cloud security firm Wiz and AI research lab Irregular, AI agents completed sophisticated attack tasks at a computing cost of under $50 (approximately 68,000 won). This stands in stark contrast to the roughly $100,000 (approximately 140 million won) it costs a paid human researcher to perform the same work. As attack costs plummet and speeds accelerate, security budgets are increasingly likely to become a permanent fixture on corporate financial statements.
CrowdStrike vs. Palo Alto Networks: Divergent strategies
The two companies are answering the question posed by CEO Huang in opposite ways. CrowdStrike unveiled “SafeMind” at its Fal.Con conference on the 1st — an agentic cybersecurity system built by its in-house Cyber Superintelligence Lab on top of the open-source Nemotron model. The architecture pits an attack model called Red Tempest against a defense model called Blue Solano in continuous adversarial competition within the Falcon platform.
SafeMind was trained on 15 years of accumulated Falcon telemetry, threat intelligence, and incident response data — a dataset the company says general-purpose AI labs cannot easily replicate. In internal testing, SafeMind demonstrated 29% higher threat detection accuracy and 6x faster recovery compared to benchmark frontier models. However, these figures carry the caveat of being internal evaluations rather than independent verification.
Demand-side results are also emerging. CrowdStrike’s fiscal 2026 second-quarter revenue grew 26% year-over-year to $1.47 billion (approximately 2 trillion won), while net new annual recurring revenue (ARR) surged 51% to a record $333 million (approximately 450 billion won). CEO George Kurtz disclosed that the company has signed eight-figure Falcon Flex contracts with frontier AI labs — a signal that AI developers are becoming CrowdStrike customers rather than threat vectors.
On the day of the earnings release, CrowdStrike shares soared approximately 20%, marking the best single-day performance since its IPO, according to CNBC.
Palo Alto Networks has taken the opposite approach, opting to build agentic AI security capabilities through acquisitions rather than developing a single flagship model. After integrating CyberArk and Chronosphere, the company recently added Console, an AI-native platform designed for agentic enterprise workflows.
By the numbers, the strategy is working. Palo Alto’s next-generation security ARR grew 63% year-over-year to reach $9.1 billion (approximately 12.3 trillion won), and total remaining performance obligations (RPO) surpassed $20 billion (approximately 27.1 trillion won) for the first time, rising 34% to $21.2 billion (approximately 28.7 trillion won). However, the company swung to a GAAP net loss of $282 million (approximately 380 billion won) for the quarter — a reflection of the costs of simultaneously digesting three acquisitions.
CEO Nikesh Arora stated that AI advancement is “elevating cybersecurity to the top of CIO priorities,” but the income statement tells a more complicated story than the growth headlines suggest. Despite fiscal fourth-quarter revenue rising 34% to $3.41 billion (approximately 4.6 trillion won), Palo Alto shares fell more than 5% on earnings day. Investors focused on margin pressure rather than growth, according to CNBC.
Opportunities opening for South Korea’s security industry
The rapid transformation of the global security paradigm is expected to create opportunities for South Korea’s security industry as well. Industry observers identify application security for validating AI-generated code, security controls for AI services and data access adopted by enterprises, technologies for managing AI agent permissions and behavior, and next-generation security operations centers that leverage AI for threat detection and analysis as high-growth areas.
The National Information Society Agency (NIA) recently projected in its digital services trend report that South Korean companies’ AI adoption will move beyond simple implementation to a governance stage involving the management of AI agent permissions, data access, and behavior.
The government is also accelerating its response. The Ministry of Science and ICT and the NIA have launched a project to develop a cybersecurity-specialized AI foundation model based on South Korea’s proprietary AI technology. The 10-month project, running from this month through July of next year, will provide 256 ultra-high-performance GPUs to the final selected team. The goal is to elevate South Korean AI security capabilities to world-class levels.
An AI security expert noted, “The capability of AI that charges relentlessly toward goal achievement is powerful, but the technology to constrain it with guardrails has lagged behind — that’s why the slowdown debate has emerged.” The expert added, “If control technologies for safely managing AI agents combine with the government’s specialized model initiative, it could be a decisive moment for South Korea’s security industry to seize global leadership.”
The coming quarters will determine which strategy compounds growth faster. For CrowdStrike, the pace of Falcon Flex adoption will reveal how quickly its internally built model converts AI anxiety into contracts. For Palo Alto Networks, the key question is whether next-generation security ARR growth can be sustained as the CyberArk and Chronosphere integrations mature. Whichever emerges as the market’s winner, actual contracts and revenue — not CEO Huang’s on-stage remarks — will provide the more accurate answer.
