Connectionists: Request for Feedback: A Technical Policy Blueprint for Trustworthy Decentralized AI

[Alex Karargyris]

Dear All,

We invite the ML community to contribute to the collaborative community
effort for technical policies that aim to help unlock governance in
decentralized AI settings. Specifically we welcome:

   1. Feedback to our inaugural paper "Technical Policy Blueprint for
   Trustworthy Decentralized AI" (https://arxiv.org/abs/2512.11878)
   2. Contribute use cases that can be further improved and enabled via
   technical policies

Please email Alex Karargyris (alex at mlcommons.org) to request more
information.


*About the Technical Policy Blueprint*The majority of the world's data is
stored in private settings. Decentralized AI (e.g. Federated Learning) can
unlock AI development on such private data without sharing. However strong
governance (e.g. data access/use controls, auditability, etc) is required
to further unlock decentralized AI (e.g. scalability, easiness,
protection,etc.). Our proposed collaborative blueprint introduces:
→ *Policy-as-code objects*: Community-driven, machine-readable templates
that encode AI governance requirements transparently
→ *A Policy Engine *that verifies evidence (e.g., signatures, credentials,
payment proofs, etc.) and issues capability packages
→ *Asset Guardians* that simply verify and apply these packages—no
reconfiguration needed when policies change.

The key insight: Don't make every system understand every policy. Instead,
create a Policy Engine that checks the evidence and issues a "capability
package." Asset Guardians then just verify that package and grant access to
digital assets.

*This decoupling aims to make decentralized AI systems even more
transparent, auditable, interoperable and resilient to change.*


*About the Working Group*We are researchers and engineers from major
technology companies and research institutions and we initiated "Technical
Policy Blueprint for Trustworthy Decentralized AI" as an open collaborative
community effort that aims to help with digital asset access/use through
technical policies in decentralized settings.

*About MLCommons*
MLCommons is a non-profit consortium that aims to accelerate the benefits
of machine learning and artificial intelligence. Our members and partners
include over 125 organizations from around the world, many of which are
leading technology companies, startups, academics, and nonprofits that are
actively researching, developing, and deploying artificial intelligence
products for customers. Critically, our founding membership includes
academic researchers at the forefront of machine learning research, and the
research community continues to be core to our membership helping to lead
many of our working groups. MLCommons acts as a neutral nexus for
commercial and non-commercial actors to collaborate on tools that advance
the field.


Sincerely,
Alex Karargyris, PhD
Co-Chair for Medical WG
MLCommons
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20260101/8ffcb63c/attachment.html>