Every founder meeting I take this year has a version of the same question. “Autonomous pentesting is here. We have been running DAST for five years.
Do we still need it?” Many CISOS are thinking “Is my scanner budget wasted now that there are agents that can actually exploit things?”
Short answer: yes, you still need DAST in 2026. The role it plays has narrowed. If you treat DAST the way you treated it in 2021, you are wrong. If you kill it because autonomous pentesting arrived, you have left the CVE floor uncovered at the worst possible moment you could.
Let’s talk some numbers
The CVE floor is bigger than ever and still accelerating. The CVE Program published 48,185 CVEs in 2025, up from roughly 40,000 in 2024, and FIRST’s mid-2026 vulnerability forecast projects around 66,000 for 2026 (The Stack, NVD data). Monthly publications roughly tripled between early 2024 and mid-2026.
The volume got so unmanageable that in April 2026, NIST formally changed its NVD enrichment policy, routing full analysis only to vulnerabilities in CISA’s KEV catalog, federal-use software, and critical software defined under EO 14028.
To put it simply, CVEs are growing exponentially, and even keeping track of them is a job NIST has decided it can no longer do in full.
No reasoning agent handles this volume efficiently, and no human triages it manually without burning heavy inference cost or heavy engineering hours. This is signature-matching work, and signature-matching work belongs to automation.
A DAST engine does it at near-zero marginal cost per release. The industry’s open complaint is that most DAST deployments throw too many false positives, and that is fair.
We vet our findings at Astra and prune aggressively, delivering zero false positives. That discipline matters regardless of who you buy from. The trick is that catching vulnerabilities early in the SDLC is the cheapest place to catch them, and DAST sitting in CI is where that economics is real.
To understand more about check out our State of Pentest Report 2026
What DAST is not for in 2026 is business logic
Scanners have never found IDOR, BOLA, broken access control, authentication bypass, workflow abuse, or chained privilege escalation. AI-powered DAST may surface some, but it cannot tackle the full set on its own.
Most of these are reasoning problems and need context about the application to act on. That layer belongs to autonomous pentesting agents that reason about the application and prove the exploit end to end, and above that, to the senior human pentesters who find the novel chained abuse no training data has seen.
So where does that leave the stack?
Keep DAST in CI, continuously, as the CVE baseline. Treat it as the layer that keeps the cheap, high-volume signature work from stealing cycles from your senior people.
Thanks to the CVE count growing day by day, a production system that looked safe after yesterday’s scan may not be safe 24 hours later. Many DAST tools in the market now offer scans that run only against newly disclosed CVEs so you can catch what landed overnight without rescanning the whole surface. We ship this as Emerging Scan in our DAST scanner.
If you are running DAST without something equivalent, you have a visibility gap that scales with the CVE curve above.
Add autonomous pentesting on top for the business-logic layer DAST was never built to see. Keep senior offensive engineers on novel chained attacks, deep source review, and the compliance-signed engagements auditors still want signed by a human.
So, do we need DAST in 2026?
Yes, as the CVE floor. The teams that get the layering right spend less and catch more. The teams that pick one layer and declare the other dead usually find out, during the incident review, that application security was always a layered problem.



