TechForge

4th November 2026

Security testing has been in a race against development speed for years. And with AI coding tools in the mix, it’s getting harder to keep up. Manual pentests and remediation cycles weren’t really designed for code that ships daily.

In early 2026, researchers used AI-assisted analysis to uncover CVE-2026-1731 in BeyondTrust remote access products. The flaw allowed unauthenticated command injection and received a CVSS score of 9.9. Attackers began exploiting it within days of disclosure.

AI-driven findings like this are real, and AI pentesting makes it practical to test continuously rather than once or twice a year. Could it close the gap between development speed and security for good?

AI pentesting investigates attack paths that scanners miss

Vulnerability scanners flag far more than anyone can fix, and much of it isn’t exploitable. Some findings are false positives, others are real but difficult to exploit in practice.

A scanner might flag an XSS vulnerability because an application reflects user input in a response. That doesn’t necessarily mean an attacker can execute JavaScript in another user’s browser, because it depends on the location of the input, output encoding and the application’s security controls.

AI pentesting can investigate the issue further. It can submit test payloads, check the responses and try another approach if the first one fails. A successful, harmless proof of concept gives developers something concrete to work with. They can reproduce the issue rather than spend time figuring out whether the scanner was right.

The most dangerous vulnerabilities may not have a CVE

CVE scanning is useful for finding known vulnerabilities in software and components. But not every exploitable weakness has a CVE. Misconfigurations usually don’t.

In 2025, researchers found 1,862 MCP servers exposed to the internet. They checked a sample of 119 by hand, and all of them listed their internal tools without any authentication. Some listed tools that could write to CRM platforms and financial databases.

An open server that only offers a public documentation search is much less concerning than one connected to a financial database. AI pentesting can help establish what someone without authorisation could actually do with that access. Can they only retrieve public information, or can they reach internal records and modify data?

Turning the pentest into part of the delivery loop

A traditional pentest usually happens at a specific point in a project, often around a major release. The testers finish their assessment, send over the findings and the development team starts working through them. Some fixes may need another round of testing, which isn’t always scheduled immediately. Meanwhile, development continues.

AI pentesting can make that process continuous. DevSecOps teams could test major releases before deployment and reassess exposed applications after significant changes. They could also retest a weakness as soon as it is fixed. In a traditional engagement, that retest can come weeks later.

That changes the pentest from an occasional security event into part of the software delivery lifecycle. Findings arrive while developers still remember the code.

AI pentesting adds a check after shift-left

DevSecOps teams already use SAST, SCA and code reviews to catch security issues during development. These checks are useful, especially for vulnerable dependencies and mistakes in the source code. But passing them doesn’t mean the deployed application is secure.

What’s thin is a check after deployment that looks at the whole environment. DAST tests running applications, but typically one at a time. It doesn’t look at how systems connect.

AI pentesting adds that check. It examines the deployed environment as a whole and tests whether the weaknesses it finds can be turned into an attack path. Shift-left reduces what reaches production. AI pentesting checks what gets through anyway.

How AI pentesting is changing the role of human testers

AI pentesting won’t replace human pentesters, but it does significantly change what they spend their time on. A Stanford study neatly sums up the new split. In December 2025, an AI agent called ARTEMIS placed second against ten professional testers on a live network of roughly 8,000 hosts, outperforming nine of them. It also produced more false positives than the humans, and it missed a critical flaw that most of them found.

That changes how DevSecOps teams organise testing. AI can run the repetitive, high-volume work quickly. That includes enumerating assets, probing endpoints and retesting after every release. Human testers stop being the bottleneck for that work. Their time goes to what takes judgement. They can focus on high-severity findings, chaining weaknesses creatively and deciding which ones to prioritise for fixing.

Developers get reports quicker, full of validated findings that a human has already reviewed. The practical model is AI for speed and coverage, with humans for judgement and accountability.

Conclusion

DevSecOps can’t afford to let security lag behind development. As AI writes more of the code, more of it ships without a person reading every line. Testing from the outside doesn’t depend on that. It checks how the running system behaves when someone attacks it, regardless of who wrote the code or whether anyone reviewed it.

For teams looking to get into AI pentesting, the place to start is the applications that change most often. That is where the gap between development and testing is likely to be widest.

About the Author

DeveloperTech News

Related

5th November 2026

4th November 2026

3rd November 2026

3rd November 2026

Join our Community

Subscribe now to get all our premium content and latest tech news delivered straight to your inbox

Popular

20296 view(s)
13518 view(s)
11755 view(s)
10339 view(s)

Subscribe

All our premium content and latest tech news delivered straight to your inbox

This field is for validation purposes and should be left unchanged.