Aikido

Aikido launches agentic pentesting for Android apps

Written by
Mario Popescu

TL;DR: Aikido now pentests Android apps. The same agents that test your web apps and APIs can now work through your APK, log into the app, and reason through it, alongside the backend it talks to, in a single assessment. Findings come back with reproduction steps and are ready for an AutoFix, the same as any other Aikido pentest.

Aikido has been running autonomous pentests against web apps and APIs since November 2025. You point it at a target, the agents think like an attacker, chain weaknesses into real exploit paths, and hand back findings your developers can fix or AutoFix. Today, the same agentic pentest comes to Android.

Mobile tends to be the part of the stack that gets tested least. A manual Android pentest is pricey, usually happens once a year (if you're lucky), and often stops at the app itself. The backend the app authenticates against gets treated as a separate job, or skipped entirely. That's a strange place to draw the line, because the app and its backend make constant assumptions about each other, and the disconnect between those assumptions is where a lot of the interesting bugs can be found.

What it does

An agentic Android pentest tries to break your app the way a real attacker would. The agents install the actual build and test it from the inside. From there, a finding only reaches your report once it's been exploited and confirmed real.

To run an Android pentest you need to provide:

  • The APK - the build you want tested.
  • The Android source repository - For now, Android pentests are whitebox only, so the code is required to run the test. Blackbox support is coming soon.
  • Test users - The agents log in the way your users do, whether that's username and password or through biometric login.

Attach your backend API too, so the agents test both the app and the API it depends on , in the same run. That way, the parts users never see get tested as hard as the parts they interact with.

Findings you can actually reproduce

Every finding comes back with a summary, a description, the risk, and the steps to reproduce it, including the exact ADB commands to run. ADB (Android Debug Bridge) is the standard way to talk to an Android app from your machine, so the reproduction steps are exactly what an engineer would type to see the bug for themselves.

You can also open the attacker agent's trace and watch how it worked. It reads through your codebase and loads its mobile testing capabilities, then tries exploiting the app over ADB to confirm the issue is real.

Pentest results come with the same report types you'd get from a web pentest. The recon report and attack plan come with diagrams, threat scenarios, and the full plan the agents worked through, including the vulnerabilities they went after and the ones they thought to be exploitable but couldn't actually exploit. 

The app is only half the target

An Android app doesn't do much on its own. The conversation between the Android app and the backend is where a lot of the vulnerabilities are. An authorization check the app enforces on the screen might not be enforced on the server. An endpoint the app only ever calls one way might behave very differently when it receives a request it wasn't expecting.

Because you can attach both the Android code and the backend API code, Aikido tests the whole path in one assessment. This means that API issues end up in the same report as the apps, instead of a separate backend engagement.

If you've used Code Security Audit to reason through mobile source before there's a live build, you’re already familiar with this part. The agents use the running app and its backend together, and confirm what's actually exploitable rather than what only looks risky based on the source code.

The Aikido you know, for Android apps

Android pentests also come with the Aikido pentesting features you already know, like AutoFix and included retests. That way, the app, its backend, the findings, the fixes, and the retests stay in one place. Run AutoFix on a pentest finding to directly open a PR with a suggested fix. 

Retesting is the one thing that works a little differently on Android. Once the fix is merged, you need to rebuild the APK and upload the new version from the assessment's overview tab, then retest. Every retest runs against that updated build, and you can retest every issue the initial pentest found.

Get started

Android Pentests are rightsized, which means Aikido's indexer sizes the assessment to your actual app, so you get coverage across the whole thing, sized to what's really in it, and you're not paying a flat fee that ignores how big or small your app is. 

This is the same Aikido Pentesting you’re used to, just pointed at a new target. The agents, the reasoning, the reports, and AutoFix are the ones you already know. Everything we mentioned is live today, so get an APK in, put the agents to work, and see what your mobile app has been hiding. iOS Pentesting is coming soon 👀!

Start an Android Pentest →

What is Android AI pentesting?

Aikido's agents install your real Android app, log in as a user, and attack it the way an adversary would, working through both the app and its backend at once. Every confirmed issue also comes with the exact steps to reproduce it.

How is it different from a traditional pentest?

A manual engagement takes days or weeks to schedule and happens once, tests a version that's already outdated by the time you get to it, then goes stale as you keep shipping even after getting the report. Android AI pentesting runs whenever you want and returns reproducible, exploit-confirmed results in hours.

How fast can I get results?

Usually within hours. Upload your APK, link to your source code, add a test user, and the agents start working through the app.

Do I need to give access to my source code?

Source code access is required, since we only support white-box Android pentests for now. With the source and the running app in front of it, the agent reaches logic flaws that black-box scanning walks straight past.

Are there requirements for the APK I upload?

The build can't use certificate pinning, and it can't have root detection, emulator detection or any RASP enabled. Those protections stop the agents from installing and instrumenting the app on a test device, so upload a build with them disabled.

What kinds of vulnerabilities can it find?

Everything you'd expect from an Android penetration test, including unsafe WebView usage, deep-link and exported-component abuse, client-side authentication issues, injection flaws, broken access control, weak session handling, and unsafe API behavior. It also catches business logic flaws and authorization issues like IDORs, cross-tenant access, and authentication bypasses by reasoning about how the app is meant to behave.

How does Aikido prevent false positives?

Findings are only reported after the agents successfully exploit and confirm them against the live app. If an attack attempt can't be validated, it's dropped and never shown in your results.

What role does AutoFix play?

Because Aikido already understands your code, AutoFix generates a targeted change for a confirmed vulnerability and opens a pull request, so your team can review the fix and merge on their own terms. To confirm the fix, recompile the app and upload the new APK to retest on the same scope.

How is scope and safety enforced?

You define which targets the agents can attack and which they can only reach. Traffic runs through a proxy that validates every request against that scope, agents start non-destructive by default, and pre-flight checks run before the assessment. If something goes wrong, testing auto-pauses and can be stopped instantly.

Can I use it for compliance or audit reports?

Yes. Every run produces an audit-ready penetration test report with validated findings, proof-of-exploit, and remediation guidance.

How does it compare to a human pentest?

Aikido's agents matched human coverage and in some cases exceeded it by exploring more paths consistently, including the deep logic flaws a manual tester usually misses. The difference is cadence: a human pentest happens once, while the agents re-run on every build.

Share:

https://www.aikido.dev/blog/aikido-launches-agentic-pentesting-for-android-apps

Subscribe for news

4.7/5
Tired of false positives?

Try Aikido like 100k others.
Start Now
Get a personalized walkthrough

Trusted by 100k+ teams

Book Now
Scan your app for IDORs and real attack paths

Trusted by 100k+ teams

Start Scanning
See how AI pentests your app

Trusted by 100k+ teams

Start Testing

Get secure now

Secure your code, cloud, and runtime in one central system.
Find and fix vulnerabilities fast automatically.

No credit card required | Scan results in 32secs.