roseway4 4 days ago

Microsoft Agent Framework has been compromised[0]. The repo is also offline[1]. Possibly related?

[0] https://socket.dev/blog/agent-framework-pypi-package-comprom...

[1] https://github.com/microsoft/agent-framework

  • magnio 4 days ago

    Without any official communications, Occam's razor means we can assume at least one admin account in Microsoft GitHub org got compromised.

    • rdevilla 4 days ago

      that's not how you use occam's razor.

      • acestus5 2 days ago

        occam's razor says you are right

    • brookst 4 days ago

      Or a contributor to both of these related repos?

  • CodeCompost 4 days ago

    Why are you assuming a compromise? Microsoft Ignite is happening and they announced a bunch of stuff. They're probably consolidating repos.

eddythompson80 4 days ago

FYI - ONNX will automatically exfiltrate data about your application with no way to opt out other than building it from source with telemetry disabled. Maybe you could block DNS too.

  • plagiarist 4 days ago

    Yikes. Another bad actor who thinks they are entitled to data.

    • fhn 3 days ago

      are you new to Microsoft?

  • numpad0 4 days ago

    Isn't ONNX itself not MS specific?

    • tadfisher 4 days ago

      They probably mean "ONNX Runtime", which has unfortunately become synonymous with ONNX-the-standard.

      • logophobia 4 days ago

        Doesn't help that alternative runtime implementations are often a bit behind the standard.

  • henryteeare 4 days ago

    onnxruntime.disable_telemetry_events() works doesn't it?

    • mscdex 4 days ago

      Same with setting `ORT_DISABLE_TELEMETRY=1` in the environment before loading the library.

      • nullpoint420 4 days ago

        Not to be that guy but I don't think this is the gotcha y'all think it is. I didn't know that ONNX had telemetry.

        edit: I seem to have misunderstood the intent of the parent comments. Apologies

        • mscdex 4 days ago

          I'm just passing information for anyone who wants to disable it without recompiling anything.

          • nullpoint420 4 days ago

            Yeah I read your tone wrong, that's on me. Sorry

        • refulgentis 4 days ago

          I don’t understand what you mean. I’ve maintained a decently popular ONNX wrapper for years, had 0 idea there was telemetry, and this is quite useful info to have, particularly because it would be a PITA to rebuild binaries right now.

          • extrano84 4 days ago

            I only found out ONNX had telemetry because one time my program (that uses it) out of nowhere started hanging on closing. Looking at the command line output I realised it was because ONNX telemetry couldn't connect to their server and it just kept retrying the connection.

            • refulgentis 3 days ago

              Good lord.

              I was doing a Dart wrapper for MXC* this week, and found it was doing telemetry, though IIRC it was forthrightly/clearly documented.

              Been on this planet long enough that I've seen plenty of MBAs saying "hoover every piece of data every time we can" and eng going along without thinking or saying anything.

              I'm still somewhat surprised Microsoft would do that in SDKs.

              * tl;dr: MS' CLI sandbox a la bubblewrap

          • snnn 3 days ago

            Previous developer here. To clarify, ONNX and ONNX Runtime are two different things. ONNX Runtime is a Microsoft project while ONNX is community owned. Previously ONNX Runtime collected telemetry for their Windows users that use the prebuilt binaries from Microsoft. No data collection for non-Windows cases.

    • snnn 3 days ago

      Previous developer here. Previously there was a bug preventing it working as expected. Someone reported the issue to us via Github issues. If the repo is still alive, you can check the issue and code. ONNX Runtime is almost fully open source. You can audit the code. Before I left, I can say for sure that if you build everything from source there would be no telemetry. However, Windows itself may still collect data, which is a separated and large topic.

  • xpct 4 days ago

    Does it use network for anything else? Or does it run a "am I online" check on startup?

    It's a bit disappointing to me that software which I expect to be fully offline, sometimes isn't.

    • pluc 4 days ago

      Oh you must not be familiar with Microsoft

  • gbdubs 4 days ago

    I don't think this is true, with an env var of `ORT_DISABLE_TELEMETRY=1 stops sending telemetry on all of the builds I tested.

    The flag/telemetry client doesn't exist on v1.28.0, and when I tested on v1.29.0, v1.30.0, v1.31.0, v1.31.1 and v1.32.0 running on MacOS, all built from source locally at those various checkpoints, I don't see the any of the telemetry infrastructure built (device ID is unset, no network queue created, no network session storage created).

    Could you provide more context on what the issue here was/is? I might be missing it.

spdustin 4 days ago

I wonder if they found new vectors that backdoor'ed models could exploit, like that discussed in the most recent Phrack [^0]

[0]: https://phrack.org/issues/73/backdoor-attacks-against-machin...

  • momojo 4 days ago

    > The ONNX Runtime (ORT) is the most widely used engine for serving deep learning models in production. It is common for people to download ONNX models from public repositories, such as Hugging Face, and to serve them to their customers without too much scrutiny. In March 2026, we noticed a parser flaw: an ONNX model can cause ORT to read from arbitrary file paths (or memory addresses) without validations.

nvader 4 days ago

We found out about this this morning, when evals started failing in our pipelines. I've been looking around, but it doesn't seem like there was any advance warning about this.

Does anyone know why or how a repository with 22k stars just went dark?

  • FesterCluck 4 days ago

    Probably being used in supply chains, and got attacked.

    • robjam 4 days ago

      That would be my guess also. About 6 months ago the azure-functions-host repo disappeared for a while (I got a concerning automated email from GitHub from my forked repo citing a ToS violation).

    • quietbritishjim 4 days ago

      If I understand your implication, that's not what "supply chain attack" normally means. That term means attacking the mechanism that delivers your packages to deliver a different, more compromised, package. If the package itself is already compromised then that's just a straightforward dependency vulnerability.

  • Neywiny 4 days ago

    Why do you think star amount has anything to do with ability to go dark?

    • tclancy 4 days ago

      From a pure lumens perspective it seems obvious.

  • binsquare 4 days ago

    it's called "Open Neural Network Exchange (ONNX)", perhaps microsoft is taking a page from OpenAI on what open meanings.

    • giancarlostoro 4 days ago

      Considering Microsoft co-developed ONNX with Facebook... I think that the more likely scenario is they took it dark after learning that it has security issues, but they should really have left a message or something, this is amateur hour tier. GitHub repos should let you leave a "Sorry we've hidden our repo, we learned about security issues" which kind of doesn't help anyone either... since now they have to find help elsewhere, and fall victim to sketchy 'forks' from other people...

      • Yoric 4 days ago

        Yeah, not sure how many people remain at Microsoft who actually know how things work.

        • giancarlostoro 4 days ago

          After those voluntary exit packages, I'm inclined to believe this. Foolish gamble on Microsoft's part.

dangoodmanUT 4 days ago

This is why we vendor everything.

You never know when it's going to vanish

  • KirinDave 4 days ago

    If it were me, I would take the hint and maybe not trust a vendored version right now?

    • cjbgkagh 4 days ago

      I believe vendoring in this context means forking it and serving it from your own repo.

      • katbyte 4 days ago

        No that’s forking, vendoring usually means keeping a local copy of a library in /vendor that is automatically committed - rust and golang have first class support for this

      • kenferry 4 days ago

        But if we think they took it down because of security issues…

  • rtpg 4 days ago

    I do wish I could pay some money for a package manager intermediary that "magically" handles things (for example defaulting to not promoting things until 2 weeks have passed), has better perf, and generally just does the right things.

    Of course I wouldn't pay _that_ much money for it.

    • rwiggins 4 days ago

      Doesn't Artifactory provide exactly that? Along with having a bunch of features that are generally the right things.

      Though I'm pretty sure their enterprise pricing is "give us a call." Maybe your last sentence was a nod to this.

    • pjmlp 4 days ago

      I don't get it, this is exactly how I have been mostly working since around 2010.

      Nothing that lands on CI/CD pipelines is able to download anything directly from Internet, rather internal repos based on Nexus, Artfactory and similar.

      Additionally everything that gets uploaded into them needs to have an approval process for the most part.

      So, not only already exists, it is an intermediate process, and you can even pay depending on the features.

      • kreco 4 days ago

        It baffles me that this is not the default.

        I mean, sure for hobby project you don't want to setup/pay some intermediate repos.

        But for anything else you just want control, if new version of an app is released, someone needs to actively bring it to the intermediate repository.

        This is tangential, but that's also why being able to import a package from a git url in some language is extremely risky (your github repo could change owners at any time). At least the code should only know about the package id, and whatever "package manager" is install should link it to a repository.

        • duskdozer 3 days ago

          The git url could point to a location that you control.

          • kreco 3 days ago

            Yes, but should be safe by default, and then you can remove some safety net for convenience, being fully aware of the trade off.

    • malikolivier 4 days ago

      Such solutions exist. Some people are building it on their own. Some people just commit all their dependencies in their repository.

      If you want it enforced for a whole team, with caching on top, that's where registry proxies come in (Artifactory, Nexus, and similar). The big ones take real effort to set up, while lighter ones may be all you need.

      Vendoring your dependencies into the repo is the low-tech option that survives an upstream repo vanishing.

      Disclosure: I work at a company which sells such solutions.

    • binocularis 4 days ago

      I’ve been using StableBuild for this sort of thing. It’s not exactly what you’re describing, but might be relevant.

      You basically point your builds at it instead of the upstream repos. Once you’ve pulled something, it keeps that version around, even if it gets deleted or changed upstream.

      Not sure if that covers what you need, but it’s worked well for me.

UncleOxidant 4 days ago

Aren't there other ONNX runtimes out there? A quick google search shows a few written in Rust: Tract, RTen, WONNX, Burn&Candle. Also, IIRC there's a Julia-based one as well.

  • antimora 4 days ago

    https://github.com/tracel-ai/burn-onnx

    I maintain this project at the moment.

    It convert ONNX models into native, backend-agnostic Burn code for inference and fine-tuning.

    • ytxm 4 days ago

      Have used it with wasm and it's lovely, but I was kindof disappointed to learn internally it's basically another serialisation format. I imagined gigabytes of generated Rust code hehe

      • antimora 4 days ago

        The graph is converted into Rust code that calls Burn APIs but the weights are converted into Burn's format, called burnpack (similar to safetensors but instead of JSON we use CBOR for weights meta data).

    • citizenpaul 4 days ago

      This looks awesome, i may be using this in prod very soon. Ive had some ONNX issues this looks to solve

  • robertknight 4 days ago

    I maintain RTen (https://github.com/robertknight/rten). It is a good drop-in replacement for ONNX Runtime for many use cases where you are running inference on CPU (there is no GPU support yet). There is no Python or NodeJS API at present, though you could get an agent to build a small wrapper quite easily.

    • robjam 4 days ago

      I've used your library ocr to read serial numbers on an IoT device on a mobile phone. It was really responsive and faster than tesseractjs in a web worker. Just wanted to say thanks for such well thought software!

  • coffee_am 4 days ago

    Yet another alternative: convert your `.onnx` file to GoMLX [0] using onnx-gomlx [1].

    For CPUs using SIMD it's significantly faster than ONNX. For NVidia GPUs, depending on the model, you can use an XLA backend. On Apple you can use an MLX backend (but it supports Neon/SME as well).

    - [0] : https://github.com/gomlx/gomlx - [1] : https://github.com/gomlx/onnx-gomlx

vrighter 4 days ago
  • coffee_am 4 days ago

    Thanks for the article, very interesting. But not likely the exact cause (maybe a variation), according to it:

    ``` Disclosure timeline:

    2026-03-07 - bug found 2026-03-10 - reported to MSRC (#109845) 2026-05-19 - patch 6a8f021b (#28408) merged 2026-05-29 - patch 77d32c3a (#28709) merged 2026-06-12 - onnxruntime v1.27.0 released, including both patches ```

  • _bent 4 days ago

    > Disclosure timeline:

    >

    > 2026-03-07 - bug found

    > 2026-03-10 - reported to MSRC (#109845)

    > 2026-05-19 - patch 6a8f021b (#28408) merged

    > 2026-05-29 - patch 77d32c3a (#28709) merged

    > 2026-06-12 - onnxruntime v1.27.0 released, including both patches

    • snnn 3 days ago

      But I don't understand why it is an issue. When you need to serve any thing via HTTP, shouldn't the server be put in chroot and/or with selinux?

gabriela_c 4 days ago

As a former (tiny patch, pre-AI) contributor, didn't they have a license that requires everybody's approval to do this?

  • Arainach 4 days ago

    No. Even if a license required all contributors to agree to change the license, there's no requirement to keep it available on Github.

Gluber 4 days ago

It seemed to have stalled in investment before i think. Performance especially for CPU only models ( there are some use cases for this ) seemed to have lagged a lot, behind what properly hand optimized code could do.

The gpu work was mostly handed of e.g to TensorRT and that also seems since the push for more general LLM models have not really improved.

these 4 days ago

Not to take away from the fact that I don't like that this was done, but in my experience development on ONNX has been pretty slow recently. I moved away from a few months back because I felt bug fixes were coming in pretty slow, despite being reported.

  • SknCode 4 days ago

    What do you use instead?

    • these 3 days ago

      I've switched over to Candle. I quite like it so far, and its significantly reduced my app size while having the same speed. The compiled nature of it means you need to implement your model in a pytorch-like API, but the compiler removes all unused ops which was a pain to do in onnxruntime.

nullbio 4 days ago

Time to back up everything offline - Github forks won't stick around either.

craigacp 4 days ago

How odd. https://onnxruntime.ai is still up, and the packages still appear to be on PyPI, Maven Central etc. Guess I won't submit the PR I was working on tonight then.

  • snnn 3 days ago

    Thank you so much for your years of dedication and contributions to ONNX Runtime Java! We truly appreciate all the time and effort you've put into making ONNX Runtime Java better!

sph 4 days ago

Microsoft, please, I can’t have popcorn every day.

  • fhn 3 days ago

    time to buy popcorn stock

anon291 4 days ago

Better delete all the open source matrix multipliers because we're making it illegal

Axodouble 4 days ago

Maybe related to the recent release of MAI? Or the Microsoft AI models?

tomrod 4 days ago

Very concerning!

matt3210 4 days ago

Someone with a copy should soft fork it

proc0 4 days ago

omg, I just built something depending on this. Hopefully the web version is still cached in NPM.

logicchains 4 days ago

Now Jensen's gone they can start dismantling the infrastructure that allows people to run models on their own hardware.

tekacs 4 days ago

So... all of these also vanished.

   microsoft/
     ├─> onnxruntime-genai                  — generative AI runtime
     ├─> onnxruntime-extensions             — custom operators
     ├─> onnxruntime-inference-examples     — inference samples
     ├─> onnxruntime-training-examples      — training samples
     ├─> onnxruntime-swift-package-manager  — Swift package distribution
     └─> onnxruntime-web-benchmark          — browser benchmark tool

(possible they were gone earlier, but e.g. onnxruntime-extensions had a push yesterday)