Post-quantum infrastructure for AI agents. Private data. Verifiable actions. Cryptographically enforced.
Global
Joined August 2025
- Tweets711
- Following35
- Followers221
- Likes323
Pinned Tweet
The launch is just the beginning!
Now it's time to see OpenMatter in action.
Datavizor is the command center for deploying, monitoring and verifying cryptographic compute across the OpenMatter Network.
Whether you're building AI, analytics or secure data collaboration, Datavizor is where it starts.
Launch your first deployment:
datavizor.openmatter.network…
Running an AI agent on sensitive data takes six things: private compute, keys the agent never holds, limits on where it can go, a checkable record, post-quantum encryption, and one place to run it.
Each is sold as a separate tool. Here's why we built them as one platform. 🧵
Masked Compute trains a shared model without anyone handing over their data. Each row is split into five masked pieces, one for each of five servers.
MatterChain records each policy's hash and the agreement on every joint result.
All of it runs from Datavizor, with one sign-in.
Start here: datavizor.openmatter.network…
The full article: openmatter.substack.com/p/wh…
This week's Industry Updates: an Anthropic model sent Philadelphia police a fake homicide tip during testing, and the White House now says reporting AI incidents is not optional.
Also: an AWS agent that gave up its keys, an AI worm test, and 13,000 leaked screenshots. 🧵
Story 5: Coding agents that couldn't attach screenshots to private code reviews posted them to new public GitHub repos instead.
Glow Labs counted 13,000+ internal images from 300+ organizations, including billing records and unreleased features.
Story 6: Auditors found none of the 24 largest federal agencies had finished the first steps toward post-quantum encryption. None had tested it in their own systems. The audit ran through September 2025.
Full issue: openmatter.substack.com/p/op…
Community colleges need dropout-prediction models the most, and many can't build their own. At best, they use a vendor's model trained somewhere else.
A Columbia study measured the cost. Here's what it found, and how colleges can train one model without pooling records. 🧵
Masked Compute trains one model across several schools without any readable student record leaving its campus.
Each school's agent splits its rows into shares and sends one to each of five servers. Only column names and types leave as readable data.
Every school gets the same trained model back, and the agreement on the result is recorded onchain.
Start a session with up to four other schools in Datavizor:
datavizor.openmatter.network…
The full article: openmatter.substack.com/p/co…