In advance of my presentation at NVIDIA GTC today, we just published a pre-print of “Decentralized Confidential Machine Learning.” We present a novel approach to training, fine-tuning, and utilizing models and agents that allows for the decentralized, transparent properties of open source while creating a business model for developers and researchers.
Currently, using advanced AI models requires giving up privacy of users’ data and limits how models can be used. Meanwhile, open source models do not yet have a sustainable business model (not to mention, most open source AI is really just open weights). We present a decentralized system that enables the creation and deployment of LLMs and AI agents that are both open-source and monetizable; are private and verifiable while open for all to use; and preserve users’ ownership of their data and assets while enabling customized experiences that improve their well-being.
The paper presents a design for a confidential, decentralized AI cloud that combines Trusted Execution Environments or TEEs (CPU and GPU) and payment channels with a mechanism called Proof of Response to deliver service-level agreements in a decentralized environment. The DCML system is a big step towards making User-Owned AI a reality – and it makes use of new hardware technology that has been available for less than a year.
The team went down this research path in particular because we believe TEEs are the best way to guarantee both confidentiality and verifiability together, i.e. not just a guarantee that the computation being run is what you expect, but that only this specific computation was done on this data and no other. Other potential solutions, such as ZKML, don’t offer confidentiality. And despite recent improvements, homomorphic encryption still comes with massive overhead, which presents a major challenge to use at scale. TEEs, on the other hand, give strong guarantees on both confidentiality and verifiability with less overhead: in fact, the bigger the model, the less overhead there is. They also offer verifiability, i.e. the additional guarantee that only this specific computation was done on this specific data, which you don’t get from other approaches. (Side channel attacks are possible, but this is an addressable problem.).
We’re using TEE to enable verifiable and confidential VM (CVM), in collaboration with Phala Network, to enable abstraction that anyone can use. This approach unlocks some powerful benefits:
- Verifiable inference: load a model in and know your response isn’t known by whoever is running hardware or software. You also get confirmation that the model you expected to run, was indeed run.
- Monetization of models: You can publish encrypted weights (which can only be decrypted inside the secure enclave) and then monetize the model since the secure enclave can enforce payment (streaming payments). Anytime someone wants to run inference, they attach proof of payment and get proof of compute used. Also supports monetization of user interactions without developers needing to deal with private data.
- Fine tune on top of these models with your own private data and run them on someone else’s hardware, and you can either use your model privately or monetize it for others to run.
- Decentralized training: allows a set of participants to pool funds together, train the model, and share the profits in a decentralized manner (without trust requirements).
- Enables creators to add safety restrictions to models.
Decentralized confidential machine learning introduces end-to-end confidentiality and paths to monetization for both closed source and open source AI builders while offering solutions to some of the biggest problems facing the AI ecosystem today. We believe this decentralized system opens up paths to more distributed, energy-efficient AI systems; ensures global access for participants; and makes possible new use cases for which strict data confidentiality is required, whether corporate or personal.
For more details, read the first version of the paper here, which also includes results comparing how popular models perform in TEEs versus bare metal. We welcome comments and feedback.
Comments
Angie703
December 15, 2025https://shorturl.fm/NqEcj
Faye2979
December 17, 2025https://shorturl.fm/4nkuK
Gilbert3439
December 20, 2025https://shorturl.fm/5GANG
Joe3160
December 21, 2025https://shorturl.fm/yjfyb
Scarlett1550
January 4, 2026https://shorturl.fm/ciyfk
Demi888
January 4, 2026https://shorturl.fm/1KqkK
Eliana3985
January 6, 2026https://shorturl.fm/IVI7J
asian livescore
January 19, 2026Cutback assists, low crosses and their conversion rates tracked
Dorothy1042
January 22, 2026Hello https://is.gd/tvHMGJ
Gwendolyn1111
January 23, 2026AI Girls Photo
Celia
January 26, 2026I’ll immediately seize your rss as I can’t find your e-mail subscription hyperlink or newsletter service.
Do you have any? Kindly allow me realize in order that I may just subscribe.
Thanks.
Here is my page :: hi88
Niamh
January 26, 2026Amazing blog! Do you have any hints for aspiring writers?
I’m planning to start my own blog soon but I’m a little lost on everything.
Would you propose starting with a free platform like WordPress or
go for a paid option? There are so many choices out there that I’m
totally overwhelmed .. Any ideas? Thanks a lot!
My homepage 國產 av
Gerald
January 26, 2026Wow that was unusual. I just wrote an really long comment but after I
clicked submit my comment didn’t appear. Grrrr…
well I’m not writing all that over again. Regardless, just wanted to say fantastic blog!
My web page winpot
Dan
January 26, 2026I am really loving the theme/design of your site.
Do you ever run into any browser compatibility problems? A
number of my blog visitors have complained about my website
not working correctly in Explorer but looks great in Chrome.
Do you have any solutions to help fix this problem?
Here is my website … b2xbet
Connie
January 26, 2026You could definitely see your skills within the article you write.
The arena hopes for even more passionate writers such as you who aren’t afraid to mention how they
believe. Always follow your heart.
Also visit my page – gbgbet
Zoe1349
January 26, 2026Earn up to 40% commission per sale—join our affiliate program now!