AI penetration testing services compared: 16 providers, sourced

AI penetration testing services are pentests where AI agents do some or all of the testing. This page compares 16 of them, plus the publisher of this site, using each vendor's own pages checked on 9 November 2026. The list is alphabetical and nothing is ranked.

The phrase "AI penetration testing" also covers services that test AI applications. That is a separate topic, covered in AI red teaming companies. Here, AI is the tester.

Delivery models and what they mean for you#

Providers in this comparison fall into three delivery models, taken from each vendor's own description.

Three providers sit between the labels. Hadrian says findings are human-reviewed but not who reviews them. Penligent is desktop software you operate yourself, with a manual mode. Penti sells AI credits and a separate pool of expert hours.

At a glance#

Prices are the vendor's published figures. A dash means the pages checked list no price. Units differ, so do not rank by amount.

Provider by provider#

Each entry gives the delivery model, scope, price, the turnaround claim and what it measures, the retest terms and a limit. "Good fit when" is our reading of the vendor's pages, not a recommendation.

Aikido#

AI agents map, attack and verify. A human-led CREST pentest is available on request. Scope covers web apps, APIs, network, mobile and AI apps. The typical pentest costs $4,000 per assessment for one application plus primary APIs, whitebox, and the pricing page says black-box and grey-box cost extra. The turnaround claim is results on the day of the run, which measures an automated run; the AI pentest page also says a few hours and, in one place, days. Retesting is free for up to 6 months. Limit: the Rightsized range is very wide, so the real price is known only after scoping. Good fit when you want a low-cost, fast automated test for audit evidence.

BreachLock#

Human-led PTaaS from an in-house team described as expert-led and AI-accelerated, plus a separate autonomous product (Breach360). The pricing page lists Standard, Extended and Extensive tiers with no dollar figures. Tests can be scoped and started within 24 to 48 hours, which is time to start, not report time. Free manual retests are 1, 2 and custom by tier, while the product text promises unlimited retesting. Limit: no sample report is published, and the two retest statements differ. Good fit when you want a human team and a stated retest count per tier.

Cobalt#

Vetted freelance pentesters from the Cobalt Core, AI agents and a separate autonomous pentest option. The autonomous web application test costs $3,500 per test as a limited-time offer that must start and finish before 31 December 2026, with findings in 24 hours. That clock measures time to first findings. Annual tiers start within 1 to 3 business days and have no published price. Retests are free for 6 or 12 months by tier, though the FAQ says unlimited during the contract term. Limit: human-led pricing needs a quote. Good fit when you want human pentesters on a platform with in-platform retesting.

Escape#

An application security platform where AI runs the pentest, with the vendor describing no human in the loop. Scope covers web apps, APIs, network and AI apps. Results arrive in minutes after a merge, deploy or schedule, which is run time for an automated test. Open issues are re-checked. Limit: no price, no units, no sample report, and the vendor says it does not test internal network lateral movement. Good fit when you want pentest-style testing on every release next to DAST.

Excession Labs (publisher of this site)#

The lab's AI testing agents run testing tasks under tool rules. Excession Labs agrees the scope in writing, checks the evidence for each finding and writes the report. It covers web applications, APIs, multi-tenant SaaS and AI applications. It publishes no prices and no turnaround. It offers retests of fixed findings and a free pentest pilot with a small written scope. A public sample report is an example built for a made-up app, not a customer report. Good fit when you want to judge the lab through a scoped pilot first.

Hadrian#

A fleet of AI agents runs the Nova tests. Findings are described as human-reviewed, but the page does not say who reviews them. Scope lists web applications. Nova starts at EUR 1,250 per test, excluding VAT, and the page does not define what one test covers. Most tests finish within 24 to 48 hours, which measures test duration after start. Retest terms are not published. Limit: no sample report, and no API, cloud or mobile scope listed. Good fit when you want a published starting price for a quick test before a launch.

HackerOne#

H1 Pentest is fully human-led. H1 Agentic Pentest pairs AI agents with human pentesters who review every agent finding. The agentic tier covers web apps and APIs. All pricing routes to sales. The product page gives no start time. HackerOne docs list the final report 3 to 5 business days after testing ends. Docs say retests are unlimited during the remediation period, typically 30 or 90 days, though older posts describe other terms. Limit: no price or sample report. Good fit when you already use HackerOne for bug bounty or disclosure.

Horizon3.ai (NodeZero)#

NodeZero is software that runs the attacks itself, and the pages checked describe no human pentesters. Flex is $15 per asset per test on a 12-month term. Core, Pro and Elite are annual subscriptions at $50, $65 and $85 per asset. WebApp Flex is $10,000 per app per test, and the page does not define the scope of one test. No turnaround is published. Limit: costs scale with asset count, which makes comparison with per-test prices hard. Good fit when you can count your assets and want internal, external and cloud tests often.

Intruder#

AI Pentesting uses autonomous agents that work from your source code and check each finding against the live app. Scope is web apps. The pricing page shows $3,500 per test with unlimited retesting, and a 4-pack at $12,000 to use within one year. The product page says $4,000 standalone, so two Intruder pages disagree. Setup takes minutes and results arrive within hours, which measures launch to report for one test. Limit: no sample report, and no human review is described. Good fit when you can share source code and want a per-test price.

Novee#

AI built on Novee's own offensive security model plus frontier models, with test plans you can review before execution. Scope covers web apps, APIs, mobile backends and AI apps. Retests run automatically after fixes and on new deployments. No prices, turnaround or sample report are listed, and the pricing page returned 404 on 9 November 2026. Good fit when you want continuous AI testing with SaaS, bastion node or on-premises deployment.

Parameter#

Autonomous AI agents run the pentest and pause for the customer's approval before chaining or escalating. Scope is web apps and APIs. It publishes three tiers with no amounts. First findings arrive in 24 hours and most runs finish on the day they start, which measures start to first findings. Free retesting is listed for Single App and inherited by higher tiers. A sample pentest report is public. Limit: no prices, and its SOC 2 report is Type I only. Good fit when you want to read a sample before you talk to sales.

Penligent#

Desktop software for Windows, macOS and Linux that you operate, with an agent mode and a manual mode. The page describes no managed human testers. Free costs $0 per month with one trial task, and Pro costs $99.9 per month for 13,000 credits. The homepage tagline says an hour replaces a week, a marketing claim with no timeline. Pro includes re-runs and re-scans. The 19-page sample report uses synthetic data and needs a free account. Limit: you own the judgment on findings. Good fit when you are a consultant who wants to run AI-assisted tests yourself.

Penti#

Autonomous AI agents run the exploits. A certified pentester can verify findings: 3 per year on Plus, 6 on Advanced, and a paid add-on on Launch. Monthly plans run $20, $300, $1,000 and $2,000 in credits, and expert hours cost $4,000 per 15 hours. Every plan includes unlimited retests. No per-plan turnaround is stated. Limit: credit cost per test is approximate, so a full web app test is hard to price, and no downloadable sample exists. Good fit when you want monthly pricing and an option to buy human hours.

Synack#

Sara, an autonomous AI agent, works with the Synack Red Team of vetted researchers. Sara Pentest starts at $5,085 per test, external web or host only. SynackST starts at $10,283 and Synack14 at $27,120. The turnaround figures are test windows: 4 to 5 business days, 5 days and 14 days. Patch verification is included in each product. Limit: the required platform subscription has no published price, and Synack and NetSPI announced a merger on 2 September 2026, so offerings may change. Good fit when you want several depth levels with starting prices.

Terra Security#

AI agents run the tests with human pentesters supervising, and certified pentesters sign the reports. Scope covers web apps, network and AI apps. The vendor says tests drop from 4 to 6 weeks to 2 to 4 hours. That is a marketing comparison, not a service level. No pricing page, retest terms or sample report are published. Good fit when you want continuous testing with a human on the loop.

TurboPentest#

Self-service AI agents run the pentest. The page does not say humans validate findings. Audit-Ready is $99 per credit for one target, Threat-Hunt $299 for up to four, and Adversarial-Depth $699 for up to eight. Tests typically complete in a few hours. A retest is a new paid run, with a free re-pentest only if the first run has zero actionable findings. A sample report is public. Limit: business logic areas are flagged for manual follow-up. Good fit when you need a low-cost external web and API test and accept automation only.

XBOW#

The vendor says no human drives the attack. Autonomous agents test and findings are reviewed before delivery, though who reviews is not stated. Scope is web apps and APIs. Pricing is usage-based with no unit or amount published. First validated findings arrive in hours, which is time to first findings, not a full report. Retests are described as free and unlimited. Limit: no price or sample report. Good fit when you run many web apps and want continuous testing with scope controls.

How we compared these services#

We read each vendor's pricing page and AI pentest page on 9 November 2026 and recorded the figures in our open dataset. We record a price only when a page shows an amount and a unit. A turnaround claim is kept together with what it measures. We did not run any of these services, so we make no claim about test quality. Outpost24 was checked, but its pages returned HTTP 403, so we left it out of this comparison. Services that use human testers only are not included. Vendor facts can change; use the correction form if you see an error.

What should you check before you buy?#

Start with what you are buying: a one-off test, a credit pool or a subscription. Then ask four things.

1. What does one test cover? Hadrian and Horizon3.ai both publish a price but do not define the scope of one test.
2. What does the turnaround clock measure? Use the questions to ask a pentest provider.
3. Is a retest free, and for how long? TurboPentest charges for a new run, Intruder offers unlimited retests, and Aikido offers 6 months.
4. Can you read a sample report? Judge it with the sample report guide.

For tools you run yourself, see the automated pentesting tools guide. For the human side of the question, see AI vs human pentesting.

Provider cards#

We show the four providers with a public sample report, so you can read before you contact anyone.