Just sent this to my boss. Felt like tossing a grenade over a fence into a party of unsuspecting people.

We don’t use ActiveStorage but Claude was able create a similar exploit in own our app in the exact same way via our own file upload library in 3 minutes simply by point Opus 5 at our site and asking it if we were vulnerable to an attack similar to KindaRails2Shell.

What a time to be alive.

The fix is pretty easy -- you should call `block_untrusted` to stop loaders like matlab from running:

https://www.rubydoc.info/gems/ruby-vips/Vips.block_untrusted

You can also set the env var `VIPS_BLOCK_UNTRUSTED`, which might be easier.

You can block or allow specific load operations, so you can limit format support to just the types you need:

https://www.rubydoc.info/gems/ruby-vips/Vips#block-class_met...

That might be even better.

There was a post on libvips.org about this a while ago:

https://www.libvips.org/2022/05/28/What's-new-in-8.13.html

And a note about it in the checklist for devs:

https://www.libvips.org/API/current/developer-checklist.html...

This post could be 10% as long:

- There was a bug with a patch

- We applied it to our clients

- There were live exploits within eight hours of the patch being released

- The Rails team had to expedite release of the technical details because POCs obviated the need to embargo

What is shocking to me is that it took eight hours.

There are so many CVEs related to upload… and basic user/group/file permissions and proper rules within the web proxy could mitigate them.

It reeks of people just writing stuff and tossing it up thinking that they’ve crafted something so great they needn’t worry. Get a good platform team.

Nice write up, Claude.

Do you have to have matlab running on your rails server for this to happen?

Not running, but supported. You can check your app with:

    bin/rails runner '
      require "vips"
      puts "ruby-vips #{Vips::VERSION}  libvips #{Vips.version(0)}.#{Vips.version(1)}.#{Vips.version(2)}"
      begin
        Vips::Operation.new("matload")
        puts "matload PRESENT - this build can reach libmatio"
      rescue Vips::Error
        puts "matload ABSENT - this build cannot reach libmatio"
      end
    '

This is from the Rails official docs for the CVE which, interestingly, they only released as an agent skill. https://github.com/rails/rails-forensics-CVE-2026-66066/blob...

An agent skill is the official distribution format for the forensics on a 9.5. I mean, I get it, anyone running a Rails app right now is pasting "am I affected" into an agent anyway, but it's the kind of thing that would've sounded like a joke a couple years ago.

Makes sense though. Agent skills are - by a mix of LLM nature and fashion - just high-quality documentation. Documentation that only gets written now, because agents are what makes docs "something immediately and directly useful for me right now", vs. "something I should write so others may benefit, someday, somehow".

Human incentives are funny.

Why would you have matlab on an external server? People don't even have a compiler on the server in this situation. Crazy.

I think the answer is no - this would affect any Rails app with default settings that uses ActiveStorage. The "Preconditions" recap at the bottom here has all the appropriate caveats: https://ethiack.com/info-hub/research/kindarails2shell-how-a...

I am not sure, but my read on the original disclosure is no. libvips itself has a variant processor for matlab v5 files, which the exploit took advantage of.

libvips also have a block_untrusted mode where it will block unsafe loaders, .mat seems to be marked as untrusted:

    vips -l
    VipsForeignLoadMat (matload), load mat from file (.mat), priority=0, untrusted, is_a, get_flags, get_flags_filename, header, load

Correct, which is how the ActiveStorage gem was patched. After this, Rails raises a Vips::Error: VipsForeignLoad exception on an attempted variant render of a malicious file. I plan on writing a technical detail post soon with some more code level details and "indicators of compromise" but this one was getting long. This is more for management to understand why wait to patch is a major issue. The discovery to active exploit attempt timeline is the story here.

This website is format is really weird for mobile, I can only read two lines of text. The rest is covered by a big banner. Im on IOS. Anybody else having this issue or is it just me?

Same thing in Firefox on Android but I used uBO element picker to select the header and nuke it. There is a link to the home page on the HTML below it and links in the footer. There is little to gain from that sticky header and much to lose.

Thanks for the heads up. The navigation header does not collapse as I never liked hamburger menus but it should be more than two lines. Works more than that on my iPhone 16. What size is yours? I will pull it up in the Firefox simulator next week and try to make it better.

We recently updated the design. This is a very old site so it has some quirks in the design for sure.

You’re welcome. The banner obscures the view almost totally and I can read only 2 lines of text and scroll through that. I have a small device, Iphone se2022 if it helps.

Android Firefox here, top nav takes up 1/3rd of the viewport

Same. So much spacing between the 3 lines. If they want to keep that huge header I'd suggest it gracefully entirely disappear when scrolling down and reappear when scrolling up.

Thank you. I will work on that.

i thought cloudflare would protect against those no?

Since the vulnerability is exploited by a crafted binary file, I think that isn't something that CF's managed ruleset is able to protect against. They have the ability to scan incoming files with antivirus, but if the exploit is small and simple and can be mutated per request, I think it's unlikely any AV would pick it up.

I agree. And unstated in this write up is the direct upload route. Even if your Cloudflare was perfect, once the attacker got the preflight they send the binary file up to S3 directly and then hit the variant route directly. The first code to “validate” the upload was the exploitable libvips code.

Cloudflare or a WAF may or may not help. These can often catch and block specific bot traffic, but not every attack payload is delivered naively. It would be part of a defense in depth. Having the underlying vulnerability fixed is critically important. For those on AWS, WAF & Shield is also very useful but at the end of the day these let legitimate traffic through, such as legitimately uploading a file that only in its contents is malformed.

Where does it say the site used Cloudflare?

DHH needs to focus on Rails again rather than Omarchy.

He’s still very supportive of Rails. Come join us at RailsWorld in Austin later this month and see for yourself!

I don't know. Austin isn't the city I fell in love with in the 90s and early 2000s. Chiefly because it's no longer full of native Texans.

I’m a native Texan, and I can deign to go to Austin for a couple days even though I live in a much bigger city in Texas which has delicious food from far more countries than Austin does. Plenty of Texans, and plenty of room for people from everywhere.

Took me a minute, but absolutely lovely.

What does DHH have to do with this? Omarchy itself isn’t known for being secure; here is a root escalation from five days ago https://hackernews.hn/item?id=49499854

The rails developers are incredibly smart and capable. They patched the exploit. The problem is that it’s too easy to reverse engineer based on the patch. They can’t do anything about that.

> What does DHH have to do with this?

DHH created Rails.