---
layout: post
title: "AI slop security engineering: Okta's nextjs-auth0 troubles"
tags: [ai_slop, security, appsec]
description: "When AI engineering fails: Dealing with hallucinations, misattribution, and broken code in an Okta/Auth0 pull request maintained by AI."
---

In October, I reported two security issues to Okta's auth0/nextjs-auth0 project, [here](https://github.com/auth0/nextjs-auth0/issues/2382) and [here](https://github.com/auth0/nextjs-auth0/pull/2381). The latter bug, an oauth parameter injection, [allows](https://techdocs.akamai.com/identity-cloud/docs/secure-audience-injection-and-the-resource-parameter-1) for a range of types of abuse, like scoping tokens for unintended services, setting `redirect_uri` and `scope` to arbitrary values to leak tokens, and so on.

The patch was simple enough, so I opened a PR:

```diff
diff --git a/src/server/helpers/with-page-auth-required.ts b/src/server/helpers/with-page-auth-required.ts
index 41af2dfe..f07046b8 100644
--- a/src/server/helpers/with-page-auth-required.ts
+++ b/src/server/helpers/with-page-auth-required.ts
@@ -196,7 +196,7 @@ export const appRouteHandlerFactory =
           : opts.returnTo;
       const { redirect } = await import("next/navigation.js");
       redirect(
-        `${config.loginUrl}${opts.returnTo ? `?returnTo=${returnTo}` : ""}`
+        `${config.loginUrl}${opts.returnTo ? `?returnTo=${encodeURIComponent(returnTo)}` : ""}`
       );
     }
     return handler(params);
```

All's well that ends well, right? Obviously, no.

---

The PR, 3 weeks later, was closed by the maintainer, an auth0 (an Okta company) employee, with the following comment:

> This change is superseded by #2413. This was done to ensure that commits are signed. Orignal contribution history has been preserved. Hence closing this PR now.

Hmm, let's take a look at _that_ PR:

| ![](/files/slop.png) |
|:--:|
| *auth0/nextjs-auth0 #2413* |

Hmm. That patch looks familiar. And who is Simen Olsen?

Pushing back on the attribution error, I replied:

>> history has been preserved
>
> no it hasn't. I don't know who "Simen A. W. Olsen my@simen.io" is but it isn't me and my commit here doesn't reference that name or email address at all. Was it ai generated or something?

Of course, the answer was: yes. It was AI slop. Just like [my previous post about gixy-ng](https://joshua.hu/gixy-ng-new-version-gixy-updated-checks) (a fun read for anybody dealing with nginx), the developer had used CoPilot to somebow generate their patches:

> Hi @MegaManSec I sincerely apologize for this attribution error.
>
> Can confirm that an AI workflow was used to created the rebased commit, which got confused with OP details.
> I've added a correction to #2413, and will ensure the changelog is updated.
>
> Thank you for calling this out, we'll make sure this doesn't happen again.

Not only did the maintainer state the above, they also used AI to generate the response! In a now-deleted comment, they clearly used some AI to respond to my complaint:

| ![](/files/slop2.png) |
|:--:|
| *auth0/nextjs-auth0 #2413's now-deleted comment* |

With the classic ChatGPT "_you are absolutely correct_", it's pretty frustrating that this developer used AI to:

1. Take my report/PR and commit it themselves.
2. Used AI to commit it, removing my attribution.
3. Used AI to "apologise" for using AI, then stated that "it won't happen again" -- (yeah right; please provide a detailed explanation how you're going to ensure that, when clearly a 1-line code change is too much for your AI to handle without breaking).
4. Refused to fix the commit to remove the invalid / AI-generated-slop details, and add back mine.

Indeed, asking:

> I would appreciate force-pushing a fix for the commit to properly include my information in the commit.

I was told that they cannot change it. That seems like a copyright infringement to me: taking somebody else's code, then changing the author's name?

What I really find the most _interesting_ is really _how_ this AI slop even came to be. I cannot find any reference to the email address "my@simen.io" anywhere online. On GitHub, the only reference to this email address is from the nextjs-auth0 PR. Simen Olsen has never contributed to any of the nextjs-auth0 repositories as far as I can tell (searching `org:auth0 author:simenandre` on GitHub), and that doesn't even seem to be their real email address. so was this some type of ai hallucination? And why? The code change was tiny. I just totally don't get it: I have literally _never_ had any AI tooling fail like this and come up with some other person's (fake) contact details. It's simply absurd; are auth0's engineers using some extremely (**extremely**) low quality local model or something? If ChatGPT failed like this for me even once every thousand times, I would simply never use it again.

In the end, at the time of writing this, the auth0/nextjs-auth0 maintainer, Tushar Pandey, who made all of these mistakes, has not fixed attribution mistake in the commit history. In addition to this, that [first bug](https://github.com/auth0/nextjs-auth0/issues/2382), which allows for arbitrary account hijacking in this software, has been fixed after 3 weeks, with new versions of the nextjs-auth0 software released, but Okta's security people stating that "unless you create a video abusing this vulnerability, we aren't going to accept this as a security issue" -- LMAO; "yeah, it's a vulnerability, we fixed in the code, it can be used to takeover accounts, but you need to _create a video_". Hilarious. That's just another case to add to my list of hilarious problems related to reporting security issue, that my next post will document.