5 Takeaways From the Autonomous Pentesting Pulse Report: Where AI Fits in Continuous Pentesting

Every security leader I talk to is hearing two messages about AI at once. Autonomous agents will soon find and exploit vulnerabilities faster and more cheaply than people can. And you shouldn't trust them to do it alone yet. Both messages are partly true, and the decisions you make about testing over the next two years depend on getting the balance right.

To help security teams assess this all important question, we’ve published unique research in the Autonomous Pentesting Pulse Report that examines how security buyers and the pentesters themselves see autonomous pentesting and where the balance lies today. We surveyed pentesters in the Cobalt Core twice, in January and September 2026, and drew on a separate Omdia survey of 400 security buyers. Together they show where the line between human expertise and AI sits today, and how fast it's moving.

The short version of the story is that autonomy is arriving faster than ever but trust in it hasn’t caught up. Continuous pentesting that blends human expertise with automation is how you capture AI's gains without betting your program on it. Here are my takeaways from our research that decision-makers need to know now.

Key takeaways at a glance

  1. AI is now everyday pentesting work. 61% of pentesters use AI frequently or extensively, up from 32% in January 2026.
  2. AI has moved from the write-up to the attack plan. Brainstorming attack paths is now AI's top use, and 59% say exploit development has shifted significantly to AI.
  3. Autonomous testing has earned a place in the program. The share ranking it least effective fell from 64% to 26%, yet 94% still rank a human-led model ahead of it.
  4. Trust, not capability, sets the boundary. Only 2% of pentesters would let AI chain an exploit into an attack without supervision.
  5. Buyers are betting on autonomy but buying proof. 94% of security buyers expect agentic AI to be critical or very important within two years,1 yet only 7% run it at full autonomy today.2

1. AI is now everyday pentesting work, not an experiment

Nearly all pentesters used AI at least a little in both surveys: 97% in January and 98% in September. What changed is how much they rely on it.

In our most recent survey, 61% of pentesters said they use AI frequently or extensively, up from 32% in January (up 29 percentage points). The share using it only occasionally fell from 56% to 34%. Two things drove the shift: the tools got better, and pentesters had time to learn them.

Pentesters have always built their own toolkits, and AI is now a big part of them. More than half (53%) have built their own AI pentesting tools, from custom harnesses to multi-agent workflows.

2. AI has moved from the write-up to the attack plan

The bigger shift is in what pentesters use AI for. In January, pentesters who use AI said it was most valuable for writing reports (55%). By September, their top answer was brainstorming complex attack paths (49%, up 13 percentage points). Report writing fell to 39%.

AI is still valuable for writing reports, but compared to nine months ago, it is not the most impactful use for AI in pentesting. Now that 79% of pentesters say report writing has shifted significantly to AI, it has simply become routine, and the value has moved upstream, into deciding where and how to attack.

The same is happening in some of the hardest work in a pentest: 59% of pentesters say exploit development and proof-of-concept creation have shifted significantly from manual effort to AI in the past six to 12 months. AI is part of the reasoning, which is exactly why the question of where the line between AI and human matters more than ever.

3. Autonomous testing has earned a place in the program, not the whole program

We asked pentesters to rank five testing models by how well each uncovers complex, high-impact vulnerabilities such as business logic flaws and chained exploits. In January, 64% ranked autonomous/AI-driven scanning least effective. By September, 26% did (down 38 percentage points), and the share ranking it as most effective rose from 1% to 6%.

That's a real change between January and September. But it doesn’t mean that autonomous testing can stand alone: 94% of pentesters still rank at least one human-led model ahead of it for this kind of work.

This is where continuous pentesting comes in. Autonomous testing is well suited to breadth and frequency: running on every release, covering more of the attack surface, and catching what changed. Human-led testing is built for depth: the business logic, chained attacks, and critical systems where context decides the outcome. A continuous program uses both, each where it's strongest.

4. Trust, not capability, sets the boundary

If AI is this capable, why won't pentesters let it run alone? Only 2% would let current AI chain a known exploit into a multi-step attack without supervision. Most (65%) would use it but verify most steps themselves.

As AI does more of the work, the human job is shifting, requiring deep expertise and judgment: 37% of pentesters now say human involvement is needed to validate, prioritize, or interpret automated results, up from 16% in January. That's the biggest change in the human role between the two surveys.

Two things hold the boundary in place today—reliability and cost. AI can produce different results on each run, and the compute it needs climbs steeply as testing goes deeper. I've written about both in The Token Tax: Why AI Pentesting Gets Expensive Fast and Why One AI Pentest Run Isn't Enough. Our research report shows how they shape what you can safely hand to AI today.

5. Buyers are betting on autonomy but buying proof

Buyers see the same tension from the other side. In Omdia's survey of 400 North American security professionals, 94% say agentic AI will be critical or very important to their offensive security posture within two years.1 Yet only 7% run AI agents at full autonomy today.2 Everyone else keeps a human involved, whether alongside the AI, approving its actions, or treating it as advisory.

What they buy tells you more than what they forecast. When buyers name the single most important factor in choosing an offensive security vendor or technology, 26% pick proven ability to reduce measurable risk. Only 11% pick depth of automation and agentic capabilities.3

Buyers are betting on autonomy but buying proof. Agentic capabilities will earn budget to the extent they demonstrably reduce risk. That's the same test pentesters apply: AI earns trust when its results can be verified.

What this means for your continuous pentesting program

The honest comparison was never human versus AI. It's an AI-augmented pentester versus AI working without one. For security leaders, that points to three things:

  • Put AI to work where it's proven. Reconnaissance, reporting, and increasingly exploit development have shifted to AI. Expect your testers and vendors to use it there, and expect faster results because of it.
  • Keep experts on the work AI can't yet verify. Business logic, multi-step attack chains, and business-critical systems still need human judgment directing and validating the work.
  • Revisit the boundary often. Pentesters’ assessment of AI capabilities and reliability have changed substantially over the last year. A continuous program, with autonomous testing for coverage and human-led testing for depth, lets you move the line as the capability and cost curve moves.

These takeaways only scratch the surface. Download the Autonomous Pentesting Pulse Report for the full analysis of reliability and cost, a stage-by-stage view of where AI has taken over in a pentest, and six recommendations for security and business leaders. To see how Cobalt brings human-led and autonomous testing together, explore our continuous pentesting program.

About the data

Pentester figures come from two online surveys of the Cobalt Core: January 2026 (n=198, fielded December 22, 2025 to January 14, 2026) and September 2026 (n=94, fielded September 14 to 21, 2026). Figures on where AI is most valuable and on pentesters building their own tools use pentesters who use AI as the base (January n=192, September n=92).

Buyer figures come from Omdia, Next-Generation Offensive Security Strategies Give Defenders the AI Advantage, May 2026, a survey of 400 IT and cybersecurity professionals in North America commissioned by Cobalt; Omdia results are not an endorsement of Cobalt. Changes in shares are reported in percentage points, computed from whole-number values.

 

Sources

1. Source – Omdia, Next-Generation Offensive Security Strategies Give Defenders the AI Advantage, May 2026. Importance of agentic AI to bolstering offensive security posture in the next 24 months. Results are not an endorsement of Cobalt. Any reliance on these results is at the third party’s own risk.

2. Source – Omdia, Next-Generation Offensive Security Strategies Give Defenders the AI Advantage, May 2026. Current level of autonomy of the AI agents used for offensive security tasks. Results are not an endorsement of Cobalt. Any reliance on these results is at the third party’s own risk.

3. Source – Omdia, Next-Generation Offensive Security Strategies Give Defenders the AI Advantage, May 2026. Single most important factor when selecting an offensive security vendor or technology. Results are not an endorsement of Cobalt. Any reliance on these results is at the third party’s own risk.

Omdia disclaimer

The Omdia research, data and information referenced herein (the “Omdia Materials”) are the copyrighted property of TechTarget, Inc. and its subsidiaries or affiliates (together “Informa TechTarget”) or its third-party data providers and represent data, research, opinions, or viewpoints published by Informa TechTarget, and are not representations of fact.

The Omdia Materials reflect information and opinions from the original publication date and not from the date of this document. The information and opinions expressed in the Omdia Materials are subject to change without notice and Informa TechTarget does not have any duty or responsibility to update the Omdia Materials or this publication as a result.

Omdia Materials are delivered on an “as-is” and “as-available” basis. No representation or warranty, express or implied, is made as to the fairness, accuracy, completeness, or correctness of the information, opinions, and conclusions contained in Omdia Materials.

To the maximum extent permitted by law, Informa TechTarget and its affiliates, officers, directors, employees, agents, and third-party data providers disclaim any liability (including, without limitation, any liability arising from fault or negligence) as to the accuracy or completeness or use of the Omdia Materials. Informa TechTarget will not, under any circumstance whatsoever, be liable for any trading, investment, commercial, or other decisions based on or made in reliance of the Omdia Materials.

Back to Blog
About Gunter Ollmann
Gunter Ollmann serves as Cobalt's Chief Technology Officer (CTO). With rich and diverse experience in cybersecurity innovation, Ollmann leads Cobalt's technology and services strategy, delivering AI-enabled offensive security solutions coupled with unmatched human ingenuity. More By Gunter Ollmann