Key Takeaways

  • Finding and weaponizing a zero-day vulnerability once took military units an entire year; AI scanners now surface zero-days daily and generate exploit kits before vendors release patches.
  • Frontier AI models detect millions of code flaws, but real-world threat data shows only 1 in 1.3 million discovered vulnerabilities is practically exploitable.
  • Patching every detected flaw creates operational gridlock, as enterprise teams face queues of up to 6 million reported vulnerabilities across their software estate.
  • Autonomous agents operate across multi-step attack chains rather than isolated software silos, making single-node targeted disruption more effective than universal patching.

The Economics of Zero-Day Remediation Broke

For two decades, enterprise security relied on a simple cycle: run a scanner, produce a vulnerability report, and dispatch engineering teams to apply vendor patches. Sanaz Yashar, co-founder and CEO of Zafran Security, spent years in military intelligence watching that math function in high-stakes environments. It worked only because offensive discoveries took months of human labor.

“There was years when I was working for military that it would take us a whole year to find a zero day,” (play from 2:08) Yashar recalled. “It would take us a very long period of time also to make an exploit kit to weaponize that zero day vulnerability in order to actually hack. Today AI scanners finding zero days every day and there is exploit kit, that weaponized vulnerability is existing before the fix comes out.” (play from 2:20)

When automated exploit creation outpaces the human engineering required to test and ship software fixes, the conventional defense playbook collapses. Enterprise software portfolios cannot shut down core applications to chase endless backlogs. Security teams that attempt to patch every flag end up spending capital to remediate theoretical flaws that pose zero actual operational risk.

Only One in a Million Vulnerabilities Matters

Frontier AI models have supercharged code analysis, but their volume creates false urgency. Security operations centers find themselves drowned in synthetic alerts that have no bearing on practical risk exposure.

“The business cannot just slow down the business and we say okay now we fix 6 million vulnerabilities,” (play from 5:46) Yashar stated. “The numbers that I tell you they are right. And the other piece of that was that out of all the new discoveries by MTOS or by other Frontier AI labs, just one out of 1.3 million are actually exploitable.” (play from 5:51)

Treating all software vulnerabilities as equivalent security priorities burns engineering runway. Yashar argues that vulnerability management must shift from exhaustive remediation to graph-based disruption. Autonomous threat agents execute across paths rather than isolated vulnerabilities.

“Agents, whether if there is a human behind it or whether they are completely working autonomous, they don't think in silos. They think in actual chains,” (play from 8:51) Yashar noted. “I'm one of them that I think the time of patch and scan all of it is completely over. It's about time that you do it surgical and you do it like very smart.” (play from 10:12)

Disabling the Single Critical Node

Because autonomous attack agents assemble composite pathways across infrastructure, defense does not require closing every open door. It requires breaking the sequence.

“The moment you find those chains still you cannot fix them all,” (play from 10:39) Yashar explained. “The smart people will find that one node that collapse the whole chain and go fix that one.” (play from 10:43)

Targeting structural choke points removes the need to maintain zero-vulnerability software estates. If an attacker cannot move laterally from an initial access point to the data repository, the existence of unpatched vulnerabilities deeper in the application stack ceases to be an immediate operational crisis. Engineering teams maintain uptime and ship revenue-generating software, while security shifts from reactive janitorial work to targeted network defense.

Why It Matters

This shift signals a major reallocation of enterprise cybersecurity budgets. Legacy scanning tools that monetize on seat counts and volume-based vulnerability detection are facing rapid margin compression as customers realize exhaustive patch backlogs provide little protection against agentic exploits. For private equity sponsors evaluating mid-market software and IT service assets, relying on legacy patch compliance metrics conceals real operational vulnerability while saddling portcos with bloated SecOps headcounts.