newsfilter
← Interviews
a16z46 min

Cyberattacks Used to Take Months. AI Can Do It in Minutes.

Ami Luttwak, Zane Lackey, Sarah Wang

Interview
  • Paradigm Shift in Cybersecurity: The industry is entering "Black October," a period defined by an unprecedented volume and severity of security incidents, including complex vulnerabilities and AI-driven attacks occurring at speeds never before seen.
  • AI Attack Efficiency: AI models have collapsed exploit creation timelines from months or weeks to approximately 20 minutes, enabling unsophisticated attackers to execute complex, targeted campaigns against startups and enterprises alike.
  • Open Source Model Convergence: The anticipated gap between proprietary and open-source AI models has vanished faster than predicted; open-source models are now capable of finding and exploiting vulnerabilities with near-parity to frontier models.
  • The "Every Day is Zero-Day" Reality: Elon Musk's recent assertion reflects the current threat landscape where AI can discover and weaponize zero-day vulnerabilities daily, a capability previously unattainable for human attackers.
  • PHP Vulnerability Case Study: The "PHP to Shell" vulnerability impacted nearly 100% of Wiz's customer base due to PHP's prevalence (estimated 70% of the internet), yet patching efforts remain inconsistent across the broader ecosystem.
  • Patching vs. Exploitation Disparity: While the time to generate an exploit has dramatically decreased, the time required to patch systems has not improved and may have increased over the last 20–25 years, creating a critical window of exposure.
  • Legacy vs. Greenfield Architecture: Fully automated, AI-native patching is deemed realistic only for new companies with modernized engineering pipelines (CI/CD, automated testing); legacy enterprises lack the necessary infrastructure to achieve similar speeds.
  • Defensive Architecture Recommendations: Founders and CISOs are advised to move beyond static code scanning to "continuous testing," utilizing intelligent pipelines that layer deterministic checks with varying levels of AI model cost and capability.
  • Red Agent Strategy: The optimal defense is now "dynamic" self-attacking; organizations must continuously simulate adversarial AI attacks ("Red Agents") to identify external exposure and unauthenticated surfaces before malicious actors do.
  • Shift in Security Team Focus: Security professionals must discontinue low-value compliance reporting and configuration checks (e.g., traditional CSPM) to focus exclusively on identifying critical risks, mitigating them, and ensuring rapid response readiness.
  • Budgetary Realities: Increasing security budgets alone is insufficient; the primary requirement is a fundamental redesign of security architecture, prioritizing isolation, sandboxing, and secure design principles over tool accumulation.
  • Targeted Threats: Startups holding AI API keys are prime targets for exploitation, with attackers capable of draining financial resources and stealing secrets within minutes of intrusion.
  • Historical Regression: The current threat environment is expected to mirror the "Wild West" era of the 1990s and early 2000s, potentially returning to an era of widespread remote code execution and worms that spread faster than patching can occur.
  • AI Superiority in Offense: AI currently possesses a unique advantage over humans in the cyber domain, outperforming human attackers in speed, complexity, and the ability to leverage zero-day vulnerabilities.
  • Infrastructure Vulnerability: Critical public infrastructure, including legacy internet-connected devices like refrigeration units, is at severe risk, prompting the launch of the "Scan for Good" initiative to assist non-profits and public services.
  • Wiz's Strategic Response: Wiz is deploying "Blue Agents" for detection and "Red Agents" for continuous offensive simulation, using multi-model pipelines to balance cost, speed, and detection accuracy for customers.
  • Y2K Moment Acceleration: The predicted "Y2K moment" for AI security, originally forecasted for 12–18 months out, is occurring sooner than anticipated due to the rapid maturation of open-source AI capabilities.