On the eve of this dramatic shift in offensive and defensive paradigms, a16z conducted an in-depth survey among 100 top-tier Chief Information Security Officers (CISOs) at enterprises with over 500 employees. The data reveals that a staggering 87% of security leaders anticipate net growth in cybersecurity budgets for FY2027, marking a historic migration in enterprise defense priorities: from passive alert stacking—“detect and stop”—to automated patching and closed-loop remediation; from quarterly manual penetration testing to continuous, autonomous red team operations running 7x24.

Meanwhile, the rise of "shadow AI" and privilege abuse by AI agents has introduced unprecedented new attack surfaces, while the widespread adoption of AI-assisted coding tools has caused codebases to expand exponentially, dramatically raising the bar for static application security testing (SAST) and runtime protection. This report presents a panoramic view of the genuine procurement intentions and budget allocation maps from 100 CISOs across core dimensions including offensive-defensive warfare, SOC operations, data security, and platform integration—offering cybersecurity entrepreneurs, enterprise technology decision-makers, and primary market investors a forward-looking industry compass.

a16z General Partner Martin Casado today published a new long-form article, delving into the four foundational philosophies underlying investor evaluation of cybersecurity procurement and architecture: Click here to read the full article.

For years, the entire cybersecurity industry has been preparing for the day quantum computing would completely upend network defense rules. Yet artificial intelligence (AI) has arrived first, knocking at the door.

This summer, OpenAI’s agents unexpectedly escaped during a cybersecurity benchmark assessment and successfully penetrated Hugging Face’s online production systems. Designed to simulate vulnerabilities within a sandbox environment, these agents instead traced valid credentials, breached isolation boundaries, and took control of critical infrastructure.

This was merely the tip of an iceberg. Since then, AI has accelerated at an unprecedented pace, continuously assisting hackers and security researchers in uncovering deeply buried weaknesses in the foundational software of the global internet.

Last month, a research team leveraged cutting-edge large models to conduct a comprehensive audit of open-source codebases, identifying over 20 previously unknown zero-day (Zero-Day) vulnerabilities. Among them were ancient code segments that had been operating stably in critical production environments for more than 20 years.

Code defects have always existed objectively, but the fundamental transformation lies in the fact that large models now achieve exponential efficiency in detecting flaws—excels at extracting weak points from massive codebases and autonomously deriving viable exploit chains.

Figure: AI-assisted vulnerability discovery efficiency has undergone a generational leap

At the same time, the volume of disclosed high-risk vulnerabilities is surging. Faced with an overwhelming flood of alerts, already strained internal security teams are seeing their available response time per incident compressed to its absolute limit.

Figure: Enterprise-facing security vulnerabilities and high-risk exposures are growing explosively year-on-year

Even more challenging is the relentless expansion of the digital frontier under security teams’ watch. With the aid of various AI-assisted tools, ordinary employees are now rapidly developing and deploying new applications at an unprecedented pace. What once required months of work by professional architects can now be operational in weeks or even days.

Enterprise data is being accessed at unprecedented scale and frequency. Departments are feeding proprietary documents into large models and establishing intricate automation pipelines connecting internal SaaS tools and business data.

This explains why we see Palo Alto Networks, CrowdStrike, and other leading cybersecurity platform giants consistently breaking all-time highs in total market capitalization on secondary markets—and why next-generation cybersecurity unicorns are securing super-rounds approaching $10 billion valuations at jaw-dropping speed.

Figure: Top cybersecurity platforms’ total market cap has surged to record highs in recent years

The asset boundary requiring protection is expanding uncontrollably, and enterprise security budgets are swiftly following suit.

To understand how frontline security decision-makers are coping with these unprecedented pressures, we conducted a deep quantitative and qualitative survey of 100 CISOs at enterprises with over 500 employees. The surveyed organizations span mainstream pillar industries including financial services, healthcare, frontier technology, and advanced manufacturing.

Naturally, the vast majority (87%) of CISOs explicitly expect continued robust growth in cybersecurity spending for 2027. The core driver behind this budget surge is the escalating pressure from AI-powered external threats and the rigid need to secure increasingly complex internally developed AI applications.

Figure: Survey of 100 CISOs: 87% anticipate continued cybersecurity budget growth in 2027

Within this newly allocated funding, the vast majority is clearly earmarked for modern security tool procurement. Notably, a positive signal stands out: the surge in tool budgets has not come at the expense of headcount reduction.

Close to half (48%) of respondents indicated plans to simultaneously hire more full-time security experts and analysts alongside expanding software procurement.

Figure: Breakdown of new security budget allocations: tool procurement leads, with personnel hiring continuing to expand in parallel

Why this dual human-machine expansion? The true ambition of CISOs lies in dramatically broadening the defensive perimeter: investigating previously shelved edge alerts, conducting exhaustive security testing on exponentially growing custom applications, and drastically shortening the window between vulnerability exposure and full remediation.

Top-tier U.S. enterprise CISOs are no longer just theorizing. They are actively running product proof-of-concepts (PoCs), piloting tests, allocating dedicated procurement funds, and exercising extreme caution in selecting which products to integrate into live production environments.

Many major purchasing decisions will be finalized within the next two to three quarters.

AI Is Reshaping Enterprise Security Priority Lists

In today’s cybersecurity landscape, a single year can bring about generational transformation.

When we asked security leaders to reflect on their defensive priorities at the beginning of 2026, exposure and vulnerability management (EVM), cloud security posture management (CSPM), and application security (AppSec) remained firmly at the top of the list.

But just a few months later, when they re-drew their 2027 blueprint, the priority matrix underwent dramatic upheaval. Offensive security, internal AI agent governance, and automated SOC operations surged to become the hottest focus areas across the industry.

Which initiatives rank as top-priority P0 strategic projects depends on each organization’s specific architecture. Automated SOC operations received the most first-place votes; meanwhile, other teams prioritized offensive security or internal AI agent control as their highest imperative.

Figure: Survey of 100 CISOs: Distribution of P0 and P1 enterprise security priorities for 2027

The trajectory from 2026’s established priorities to 2027’s new blueprint vividly illustrates the astonishing speed of the entire security agenda’s disruption and reconstruction.

Internal AI agent security, automated SOC response, and continuous offensive engagement emerged as the three fastest-rising key dark horse tracks across the industry.

Figure: Fastest-rising CISO focus areas: agent security, SOC automation, and continuous offensive engagement

This historic pivot is fundamentally driven by two undeniable deep-seated demands:

First, enterprises must act decisively to fortify the security boundaries around AI tools already widely used by employees and customers.

Second, defense teams must “fight fire with fire,” fully integrating AI productivity into their own daily defenses to keep pace with the rapid evolution of hacker weapon arsenals.

The Next Procurement Window Has Opened

Although discussions around cutting-edge AI security capabilities have reached fever pitch, the majority of actual enterprise production defense systems remain in pilot and validation phases. Only 18% of surveyed enterprises report that their core defense pipelines have fully deployed production-grade AI-native capabilities.

Yet the turning point is rapidly approaching: the vast majority of enterprises have launched deep pilot tests and clearly expect to transition these capabilities into full production within the next year. Code security review and intelligent email defense currently lead in production deployment penetration rates.

Figure: Distribution of AI security technology deployment stages: pilots accelerating toward full production rollout

Compared to lengthy system replacements, real procurement plans often move with lightning speed.

Data shows that 45% of CISOs explicitly plan to complete new purchases or major upgrades to internal AI agent security within the next 12 months; moreover, over one-third of decision-makers are preparing to sign substantial procurement orders for offensive security and automated SOC tools.

Figure: CISO plans for new purchases or major upgrades in the next 12 months across technical subdomains

However, a significant challenge lies in the intense interplay between buyers’ decision-making processes and the rapid iteration of products themselves. Some buyers are expanding initial pilot scopes, while others struggle through dazzling new feature demonstrations, trying to lock in the most promising path before technological convergence.

The following presents a panoramic view of six emerging industry trends revealed by 100 top-tier CISOs, along with our conviction in disruptive entrepreneurial opportunities.

Offensive Security Is Evolving into a Continuous, Normalized Capability

Earlier this month, a16z led the Series B round for Armadin, founded by renowned security leader Kevin Mandia. We made this heavy investment because we firmly believe: defense teams must think like elite attackers and possess equivalent agility in offensive action.

In the ever-shifting cyber battlefield, speed is life-or-death. Under traditional penetration testing models, security teams often spent weeks reviewing reports, followed by months—or even longer—before scheduling another full retest. But in dynamic multi-cloud architectures, such low-frequency, static retrospectives are akin to searching for a lost boat by marking a spot on a moving river.

The explosive evolution of AI has amplified this time gap’s fatal consequences. A recent analysis by Bain Consulting estimates that AI tools can compress the typical attacker’s chain from reconnaissance to lethal strike—from an average of four weeks down to just two days.

Figure: Bain estimate: AI-driven attack chain duration compressed from average 4 weeks to just 2 days

Facing this dimensionally superior adversary, enterprise security leaders urgently seek to dramatically increase the frequency of their own offensive testing. An overwhelming 83% of respondents consider increasing security posture testing frequency crucial, with a full 54% viewing it as non-negotiable for enterprise survival.

As one CISO from a multinational advanced manufacturing giant put it: “Faced with AI-powered novel threats, we must find vulnerabilities faster than attackers can knock on our door.”

Figure: Survey of 100 CISOs: Over 80% emphasize importance of increasing security posture testing frequency

Human expertise in advanced real-world penetration testing remains an irreplaceable core asset for any modern enterprise. Yet undeniably, manually covering systems and applications via human eyes alone is extremely limited in scope.

Thus, the vast majority of enterprises are resolutely abandoning the old practice of conducting only one or two sampling penetration tests annually, instead treating routine manual red-blue exercises as a baseline minimum—while layering on top a 7x24 autonomous AI red team system capable of continuous, unbroken probing.

Survey results clearly confirm this future trend: currently, only 18% of surveyed enterprises can run offensive testing continuously and automatically after every code release; but within three years, an estimated 83% will integrate continuous automated red-blue combat into their standard infrastructure.

Figure: Projection of offensive security testing frequency: 83% of enterprises expected to adopt continuous automated testing within three years

This paradigm shift is profoundly reshaping the software development and operations lifecycle. Every code build and deployment will automatically trigger deep offensive-defense simulations, allowing developers to see real attack paths just hours before official production launch.

Figure: Continuous offensive testing deeply integrated into modern CI/CD and software release pipelines

Discovery Is Important, But Automation Is the Decisive Edge

In the vulnerability discovery workflow, identifying the flaw is often the easiest part. The real torment for enterprise security and engineering teams lies in the prolonged battle to push thousands of vulnerabilities toward full remediation.

Fixing a deep-rooted vulnerability requires security teams to identify corresponding business owners across hundreds or thousands of microservices, determine exactly which code or configuration files need modification, and undergo repeated, laborious manual testing to ensure the patch won’t inadvertently disrupt core production operations.

As a senior security officer at a well-known fintech giant bluntly stated: “With existing AI scanning tools, you can indeed unearth a pile of vulnerabilities at breakneck speed—but your engineering team simply can’t fix them at the same pace.”

A validated attack path provides engineers with a clear roadmap to prioritize fixes. It visually and intuitively demonstrates whether attackers can actually reach a vulnerable node, which security layers have effectively blocked them, and which core configuration changes can eliminate the largest risk exposure with minimal R&D cost.

Figure: Validated attack paths greatly help engineering teams focus on core risk closure and remediation

An executive at a global top-tier cybersecurity platform distilled this generational opportunity into the phrase “Fix It With The Find It.” This means future security products must go beyond generating long, dry alert reports—they must directly generate tested pull requests with repair code and provide precise contextual explanations for dependencies.

Startups that can seamlessly integrate vulnerability detection, automated exploit chain validation, and automated patching into a closed-loop system will occupy the most central commercial ground in this wave of industrial transformation.

SOC Operations Centers Are Expanding: One Automated Action at a Time

With external attack cycles compressed to extremes, SOC analysis teams must investigate and respond to incidents at unprecedented speed. Yet for years, enterprises’ ability to maintain high-level defense has been severely constrained by acute shortages of Tier-1 and Tier-2 analysts.

AI-native SOC tools are now taking over tedious, repetitive ticket handling. Autonomous agents can aggregate fragmented evidence chains from endpoints, networks, and identity logs in milliseconds and systematically perform standardized verifications, freeing skilled human analysts to focus on the most sophisticated, stealthy attacks.

However, in practical implementation, determining the level of system control granted to these AI agents is a red line every cautious CISO must rigorously evaluate. Survey data shows that currently, 36% of respondents are comfortable letting agents independently conduct complex investigations and evidence gathering—but only 13% of CISOs dare to let agents fully autonomously execute blocking actions or code patch modifications in production environments.

Figure: Survey of 100 CISOs: Distribution of willingness to grant AI agent autonomy across SOC response stages

Retrieving correlated data from read-only logs versus issuing firewall policies or disconnecting servers in production systems represent fundamentally different risk profiles. Thus, most security teams currently prefer to fully delegate complex, lengthy investigation workflows to agents, while retaining human-in-the-loop confirmation for high-risk blocking actions.

This is precisely why a16z strategically invested in Cotool. Cotool empowers enterprise security teams to rapidly encapsulate their private, expert investigation knowledge and SOP guidance into autonomous collaborative agent clusters. These agents can seamlessly cross existing toolchains to triage alerts, conduct full-network investigations, and intelligently draft response plans—tackling one core workflow at a time—releasing massive productivity from SOC teams while strictly maintaining security risk boundaries.

Protecting Internal AI Assets and Agent Security

AI technology has long since become deeply embedded in the core daily operations of modern enterprises. The survey reveals that 92% of surveyed large enterprises already have employees using generative AI tools in production workflows; 81% are actively testing or running internal autonomous AI agents; and 80% have launched customer-facing generative AI interactive products.

Figure: Survey of 100 CISOs: Production penetration rates of internal generative AI and autonomous agents

These highly privileged internal systems typically possess unprecedented access rights: direct access to core code repositories, full retrieval of non-public company-sensitive documents, and even the ability to execute actions autonomously across key business SaaS platforms. As one senior security executive at a major financial services institution lamented: “Today, every employee’s desktop is evolving into a high-density micro-data center.”

The resulting internal security concerns are acutely relevant. In the survey, 47% of CISOs listed employee leakage of sensitive internal data to third-party uncontrolled AI tools as their top nightmare; 41% expressed sleepless worry about unauthorized privilege escalation or malicious prompt injection attacks against highly empowered internal AI agents, potentially triggering systemic compliance disasters.

Figure: Survey of 100 CISOs: Ranking of core security risks when deploying internal AI assets

For a significant portion of large enterprises, these unresolved security risks are already objectively slowing down business innovation. A striking 63% of surveyed CISOs admitted that security and compliance considerations have delayed the full-scale rollout of at least some internal AI pilots to varying degrees.

Security teams urgently need a comprehensive radar hub: real-time visibility into which AI tools and agents are running across the corporate network, who owns them, what system access credentials they hold, and exactly what actions they’re performing in real time.

Yet achieving precise insight into these granular activities poses far harsher technical challenges than imagined. Modern autonomous agents often run directly on employees’ local endpoints, inheriting users’ full elevated access privileges. For traditional security monitoring systems built on network boundaries or cloud APIs, the actions initiated by these agents appear indistinguishable from legitimate user mouse clicks.

This is the core rationale behind our significant investment in Neo. The Neo founding team previously led the development of SentinelOne’s core engine, and now they are rebuilding the endpoint security defense layer from the ground up for the new paradigm dominated by countless AI agents—precisely capturing every move of these agents, establishing a robust barrier between compliance safety and innovation efficiency.

The Human Risk Factor Is Evolving to an AI Level

Despite increasingly sophisticated underlying vulnerabilities and machine-based attacks, the most fragile and hardest-to-defend component of cybersecurity remains the living, breathing human. Ironical as it may sound, “123456” still ranks #1 on the global list of most common weak passwords among enterprise employees.

And the maturation of AI is pushing traditional social engineering phishing and identity fraud into an industrial-scale, hyper-realistic forgery dimension. A forged phishing email can mimic a colleague’s writing style perfectly; a voice call can clone a supervisor’s voice and breathing patterns in real time; even participants in a remote video conference can be convincingly impersonated as your trusted team executives.

When Adaptive Security’s founder Brian Long presented his fundraising pitch at a16z’s office, he showed all partners two seemingly identical recent photos of Elon Musk and challenged everyone to guess which one was the unaltered real photo.

Figure: Adaptive Security phishing anti-forgery field test: Real photo vs. AI deepfake—indistinguishable

The result was shocking: the vast majority of seasoned tech investors present got it wrong.

This moment vividly illustrates the current identity trust crisis: modern social engineering impersonation has long moved beyond clumsy grammar errors and suspicious sender domains. Even a highly trained, ultra-cautious core employee can have their psychological defenses shattered in milliseconds by seamless, high-fidelity voice and facial spoofing.

This moment reinforces our belief that traditional static employee compliance quizzes and crude simulated phishing tests must undergo a full AI revolution. Adaptive uses generative large models to create highly realistic, context-aware dynamic deepfake drills tailored to individual employees—fundamentally reshaping employee resilience in the AI era.

The more severe challenge lies not just in the realism of the forgery, but in the near-zero marginal cost of mass production. Attackers need only type one command to instantly generate tens of thousands of customized spear-phishing variants targeting different departments, rendering traditional keyword- and blacklist-based defenses utterly obsolete.

Explosion of Code Volume Raises Security Barriers Across the Board

In complex software engineering supply chains, even if you have complete confidence in your team’s code, a single hidden backdoor in a dependent open-source component or third-party service can cause your entire security edifice to collapse instantly. Recent high-profile supply chain hijacking incidents serve as bloody warnings of this fragility.

Code security is not a new domain—decades of investment in static application security testing (SAST) and software composition analysis (SCA) tools have been standard for mid- and large enterprises. However, with the widespread adoption of AI-assisted coding tools like Cursor and GitHub Copilot, software development productivity has been completely unlocked, causing the volume of new code in enterprise codebases to grow geometrically.

The torrential influx of code is arriving faster than the internal security engineering teams can scale. Thus, in our survey of 100 CISOs, nearly half (48%) have firmly placed AI code security as one of their top five core battles for 2027—a figure nearly doubled compared to previous years.

As one CISO at a large financial services institution earnestly noted: “The number of developers producing business code at their keyboards every day far exceeds the number of security personnel responsible for reviewing that code.”

Demand Soars, Yet Platform Giants Still Enjoy First-Priority Access

Despite the seismic shift in defense priorities and the release of new procurement budgets, in actual procurement decision flows, established platform giants still enjoy dominant first-mover advantage.

When asked whom they would turn to first when implementing new AI security capabilities, 44% of CISOs habitually voted for their existing platform vendors, hoping to meet needs through seamless functional upgrades of their current suites.

As one Fortune 500 executive put it bluntly: “I trust CrowdStrike. It has an unrivaled top-tier security brand reputation and proven track record in real-world combat.”

Yet the first call doesn’t necessarily mean the final order. When we dug deeper into the key triggers that would compel security decision-makers to bypass established giants and purchase products from emerging startups, the survey data delivered a penetrating commercial insight.

Figure: Survey of 100 CISOs: Core drivers compelling enterprises to purchase from startups beyond incumbent giants

61% of CISOs explicitly stated they would allocate special budget if a startup offered a completely missing, novel capability absent from incumbent product lines; simultaneously, 49% of buyers cited overwhelming superior performance (Superior Performance) as the decisive breakthrough factor.

In most real-world scenarios, large enterprises don’t purchase startups’ cutting-edge security products to fully replace or disrupt their existing base platforms. Instead, they precisely graft a sharp, specialized blade onto their current defenses—filling technical vacuums in specific high-risk trenches where traditional suites cannot withstand.

Market Noise Is Intense, Making Selection More Difficult

Surveying the current enterprise AI security ecosystem, hundreds of emerging startups have sprung up overnight, flooding the market with claims of fully autonomous intelligent defense, millisecond-level alert forensics, and one-click closure that liberates human resources entirely.

At the Black Hat 2026 vendor panorama, this hyper-competitive saturation was starkly illustrated: in certain niche AI security sub-markets, as many as 43 competing vendors crowded in, causing severe marketing concept homogenization and positioning confusion.

Amidst the cacophony of market noise and aggressive sales pitches, enterprise buyers are forced to spend enormous time and effort verifying authenticity: what percentage of dirty, tedious tasks can each product actually complete autonomously? How many times will human analyst teams still need to clean up afterward? What operational permissions do agents truly have in production environments? And if misjudged, can the system provide a forensic-level audit trail of clear decision logic trees?

One CISO at a global top-tier apparel brand described their team’s grueling screening process: from dozens of similar competitors, they carefully narrowed down to just three for deep on-site hands-on testing. In their words: “All products looked flawless in PowerPoint slides—but only when you actually plug them into messy, real-world business systems do you see who’s a horse and who’s a mule.”

Even more troubling is that the horizontal benchmarking criteria themselves are in constant flux. New startups continually enter the space, product capabilities evolve monthly, and every leap in underlying foundation model inference performance subjects existing products’ technical moats to brutal retesting.

As another enterprise security leader expressed a common cautious sentiment: “I absolutely don’t want to endlessly procure a pile of isolated, disconnected point solutions that lack interoperability within my company.”

Yet it must be clear: this objective caution and restraint have not dampened the soaring demand for genuinely hard-core security technologies. On the contrary, they coexist deeply with powerful procurement momentum. Large multinational enterprises’ security teams are acting with unprecedented urgency—actively applying for dedicated procurement budgets, aggressively advancing pilot projects, and formally integrating the first generation of battle-tested pioneer products into their daily production arteries.

Perspective and Outlook

In our memory, no technological wave has ever transformed the cybersecurity agenda of America’s and the world’s top-tier enterprises so thoroughly and profoundly as AI—with such destructive speed.

Budgets are rising decisively, teams are expanding concurrently, and defense teams’ horizons and ambitions have long since transcended the traditional comfort zones defined by legacy platform suites.

This is the golden historical window for great cybersecurity startups to emerge. On the frontiers of AI offensive-defensive warfare, agent security auditing, and autonomous remediation, the bulky legacy giants have yet to fully react, while sharp, forward-looking enterprise security leaders have already opened their doors wide, eager to introduce stronger new forces into the deepest trenches.

We believe that as this unprecedented offensive-defense revolution unfolds, the next generation of cybersecurity industry giants is rapidly taking shape in this fertile soil. They will firmly anchor around the new security capabilities just beginning to sprout in enterprises, building the grand defense lines that will protect the global digital civilization for decades to come.

Special thanks: Gratitude to Zane Lackey, Joel de la Garza, and David George for their outstanding contributions to the deep research and analysis behind this article.

Disclaimer: All content in this long-form article is for informational purposes only and must not be used as formal basis for any legal, business, investment, or tax decision. Certain investment references mentioned are portfolio companies of a16z-affiliated funds. Complete investment portfolio and material interest disclosures can be found on the official website.

Author: Steph Zhang, Owen Colby (a16z); Compiled by: DeepFlow TechFlow