We're building open-source tech that helps app builders & researchers get answers from data without direct access to it. Join us on slack → slack.openmined.org
- Tweets 729
- Following 0
- Followers 10,626
- Likes 2,109
Most external AI evaluation programs hit the same ceiling: the access model wasn't designed to scale.
PySyft can split evaluation into three roles. An embedded evaluator writes a reusable job against real model assets. An internal reviewer approves it. External researchers receive filtered outputs without ever touching the underlying data or going through a new approval cycle.
Our approach: openmined.org/blog/scale-emb…
1
4
24
81,859
Open weights don't make an AI system open. Whoever runs the routing layer above them decides which model users get, what it costs, and where your data goes. Is that layer open?
1
4
428
OpenMined retweeted
One of the main blockers to external oversight is *access infrastructure*. Please come join @OpenMinedOrg and help us solve this critical problem.
You can work on projects like this most recent one from @GoogleDeepMind
deepmind.google/blog/pilotin…
We're hiring.
openmined.org/careers/#open-…
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
7
6
66
13,600
OpenMined retweeted
Big Tech’s power is underwritten by their near total capture of our data. Tech companies capture information on everything: what we like on Instagram, the places we visit, what we do—even how we work and the culture we help create.
All that data is fueling hyper-addictive digital products and AI systems that threaten to erode or displace human relationships, social trust, meaningful work, and agency.
The good news: there’s a fix 👇 h/t @FamStudies
Nearly all data fueling the digital economy and AI is a co-produced good.
Governance of the systems and outputs it produces should be shared by co-producers, especially families and communities.
Our newest brief, co-authored by @RealDCochrane, breaks this down into three parts.
Part I argues that data is a co-produced good. Yet, control over its collection and use is concentrated in the hands of tech firms.
Part II makes the case for why both individual data rights and top-down data privacy regimes often fail to vindicate privacy, control, and economic interests.
Part III proposes a new class of informational rights.
13
23
14,933
PySyft just hit 10K GitHub⭐️ stars. Thank you to everyone who has contributed, built with PySyft, and supported the project over the years.
Long before AI governance became a mainstream conversation, the OpenMined community was building tools that let organizations collaborate on sensitive data without exposing it.
That work continues today.
github.com/OpenMined/PySyft
1
15
109
36,453
After nearly a decade of research and development including contributions from over 400 contributors from around the world, we are pleased to announce that PySyft has been used by @GoogleDeepMind, @AVERIorg, Singapore AISI, and @MLCommons to facilitate the world’s first double-blind evaluations of a proprietary, frontier class AI model.
In an industry first, we’re piloting double-blind evaluations for frontier AI.
By creating a secure environment where neither test prompts nor model weights are revealed, we can ensure external safety and performance evaluations of our models remain private, robust, and trustworthy. → goo.gle/3St2xan
3
18
38
2,850
...and we're looking for people to help us continue to build.
We're hiring for a Staff QA Software Engineer and a Staff Software Engineer. Apply here: 👇openmined.org/careers/#open-…
3
246
OpenMined retweeted
Today, we released our work on double blind eval of frontier models using PySyft. To the best of my knowledge this is the first of its kind. Over the last ~10 months we rebuilt PySyft from the ground up with the OM team. Still a lot to do but very happy with these results!
In an industry first, we’re piloting double-blind evaluations for frontier AI.
By creating a secure environment where neither test prompts nor model weights are revealed, we can ensure external safety and performance evaluations of our models remain private, robust, and trustworthy. → goo.gle/3St2xan
1
6
16
2,114
OpenMined retweeted
I'm exceptionally proud to have been a part of this pilot. And after 9 years of R&D, massive congratulations to the @OpenMinedOrg team on reaching this milestone with #PySyft.
4
4
27
1,109
We're hiring a Head of Recruiting at OpenMined. We know the best candidates in AI and privacy aren't always looking. This role is for someone who finds them anyway and runs every pipeline from sourcing to offer.
Check out this role and others: openmined.org/careers/#open-…
5
12
1,442
Last month, fusions overtook every frontier model on DRACO.
Open-source AI vs. closed-source AI may be the wrong debate.
Our latest research points to model ensembles outperforming individual frontier models.
@iamtrask explains why
1
3
9
751
Costs, stated plainly: the fusion runs $0.296/task (aug_2026_trio in the figure attached). GPT-5.6-sol solo scores 58.1% at $0.056/task.
Solo models remain the efficiency play. Fusions buy accuracy that no single model reaches at any price. That capability only exists as a fusion.
1
2
107
Costs, stated plainly: the fusion runs $0.296/task (aug_2026_trio in the figure attached). GPT-5.6-sol solo scores 58.1% at $0.056/task.
Solo models remain the efficiency play. Fusions buy accuracy that no single model reaches at any price. That capability only exists as a fusion.
1
19