hussh at
AI Summit @ Stanford
A private agent should belong to the person at the center. Come see what that feels like—and choose your own next step.
- When
- July 30 – August 1, 2026
- Where
- Stanford University · Stanford, California
- Rooms
- McKenzie Room · STLC 111 · NVIDIA Auditorium
One good conversation. A next step they choose themselves.
We are hiring, here, this week
We are opening a research lab in the Bay Area, and putting five million dollars behind the protocols.
Two protocols decide whether a personal agent can be trusted. PCHP — the Personal Consent Handshake Protocol — puts a verifiable receipt behind every access to a person's data, and it runs in the product today. hu_ssh is the second, and it is a published design rather than shipped code; we say so plainly because a specification that overstates itself is worse than no specification.
Turning both into standards other people can implement is slow, exacting work: real specifications, reference implementations, conformance suites, and the policy engagement that makes any of it count with a regulator. We have committed five million dollars to that effort and to the 🤫 Private Agent One research roadmap behind it.
We would like to do it alongside the faculty and students here. That is an invitation we are making, not a relationship we are announcing — we will name a collaboration once there is one to name.
Roles open now
Protocol Specification Engineers (PCHP, hu_ssh)
$200k – $280k
You have written a specification other engineers implemented from.
Policy & Regulatory Lead
$150k – $240k
Part of the job is telling us, in public, what we cannot claim yet.
Protocol Developer Relations
$150k – $215k
You own a developer's first hour with our protocols.
AI / ML Researcher-Engineers
$150k – $215k
World models, memory, and the evals that keep an agent honest.
Systems & Data Researchers
$150k – $215k
On-device, edge, efficient, sovereign.
Forward-Deployed Engineers
$120k – $185k
You go where the customer is and stay until it works.
Indicative US base ranges, not offers. Equity for every full-time teammate. Final compensation depends on level, location and experience.
For faculty, alumni and trustees
Four problems we cannot solve, and a process that lets you outvote us on them.
We have published two protocol specifications, a 252-scope consent registry, a removal contract, and a proposal process modelled on the one the Model Context Protocol uses. What we do not have is a single enhancement proposal from anybody outside this company, or a single maintainer who does not work here. Those two numbers are both zero and they are printed on our own governance page.
That is the ask. Not sponsorship, not a letter of support, not a logo. The specific, unglamorous thing a standard needs early is people with standing who will read it adversarially and write down where it is wrong.
Faculty and PhD students
Take one of the four open problems as a project.
KV-cache representation across model families, revocation reaching an in-flight burst, attestation under a compromised owner device, and consent a person later regrets. None are solved, all are thesis-shaped, and a negative result published against us is worth more to the protocol than a confirming one.
Read the open problemsAlumni in industry
Implement it somewhere real, and tell us where it broke.
Three production implementations by organisations other than us is the published trigger that moves the specification to a neutral foundation, with no seat reserved for us. An implementer's bug report outranks a maintainer's opinion in every review we run.
Read the specificationTrustees and institutional leadership
Send an engineer, not a delegation.
Membership is individual and never corporate — no organisation holds a seat, which is exactly what makes participation safe for any organisation. Nothing to sign, no fee, and nothing a contributor does obliges the institution behind them.
Read the governanceA note on how we are asking. This page names no professor, claims no review that has not happened, and asserts no relationship with the university. That discipline is owed to Luiz André Barroso, who was Manish's mentor and his last boss at Google, and who taught a generation to treat the datacenter as a single computer — and to be unembarrassed about saying plainly what was not yet known. He died in 2023. Nothing here is endorsed by him, his family, Google, or Stanford. It is gratitude, and an attempt to ask for help the way he would have.
Your information stays your decision.
Our event team can show you the product and share a link. They do not need to create a profile for you, harvest a contact list, or make your decision for you.
- Open a link on your own device.
- Sign in with the identity you already use.
- Control any future sharing with consent.
- Ask to claim, correct, or remove a directory listing.