PROTECT
Establish a security boundary around the workload and its data.
CONFIDENTIAL COMPUTE RESEARCH AND INFRASTRUCTURE
COMPUTE ANYWHERE.
TRUST NOTHING.
Reveal naught.
Building infrastructure for workloads that should not have to trust the machines executing them.
WORKLOAD ≠ MACHINE
01 / THE ASSUMPTION
TODAY
THE MACHINE SHOULDN'T
NEED YOUR TRUST.
Today, sensitive computation usually requires trusting whoever owns or operates the machine.
That assumption doesn't scale to a world where compute comes from everywhere.
Your workload should remain protected—even when the machine isn't yours.
SECURITY BOUNDARYWORKLOADTARGET: DATA CONTAINED
02 / THE PRINCIPLE
THE MACHINE GETS
NAUGHT.
Naught is working toward execution models in which infrastructure can run protected workloads without gaining recoverable access to customer data.
The applicable guarantees will depend on the workload, platform, and threat model.
01DATA
02PROTECTED
WORKLOADBOUNDARY MODEL
WORKLOADBOUNDARY MODEL
03UNTRUSTED
COMPUTEMACHINE OUTSIDE TRUST BOUNDARY
COMPUTEMACHINE OUTSIDE TRUST BOUNDARY
04TARGET: PROTECTED
RESULT
RESULT
03 / ARCHITECTURE DIRECTION
A SECURITY BOUNDARY
ATTACHED TO THE WORKLOAD
PROTECT.
EXECUTE.
LEAVE NOTHING.
Naught explores what changes when the trust boundary moves away from the provider and stays with the computation.
EXECUTE
The aim is to run compatible workloads without making the selected machine the foundation of security.
LEAVE NOTHING
The target: no recoverable customer data after execution, within a defined threat model.
04 / FUNGIBLE COMPUTE
THE LARGER THESIS
COMPUTE IS
BECOMING A POOL.
GPUs sit across clouds. Servers sit in datacenters. Compute exists at the edge. Compute capacity often sits idle.
Shared compute becomes dramatically more useful when trust in the machine is no longer the prerequisite.
MANY MACHINES. ONE WORKLOAD-BOUND SECURITY MODEL.
05 / EXECUTION NETWORKS
SECURITY FOLLOWS THE WORKLOAD
SECURITY FOR A WORLD
OF LIQUID COMPUTE.
Execution layers can route workloads across heterogeneous compute infrastructure.
Naught explores the complementary question:
What if the workload didn't need to trust whichever machine the router selected?
Jungle Grid is one related execution-layer vision. Naught remains a standalone security and infrastructure research direction.
SECURITY FOLLOWSWORKLOAD
EXECUTION LAYER selects available compute
GPUCLOUDEDGESERVER
06 / FOR BUILDERS
AN OPEN ENGINEERING DIRECTION
BUILD FOR MACHINES
YOU DON'T TRUST.
Interfaces should make the security boundary visible, legible, and testable. The workflow shown here is illustrative—not a released product or CLI.
Follow the project$ naught run workload
protecting workload...
✓ execution boundary established
✓ compatible compute selected
✓ protected result returned
target host-retained customer data: none
07 / RESEARCH & OPEN PROBLEMS
NO FALSE FINISH LINE
THIS IS A HARD PROBLEM.
THAT'S THE POINT.
Naught is an engineering and research effort toward safer computation on infrastructure outside the user's trust boundary.
Some workload classes benefit from technologies available today, subject to explicit threat models and platform assumptions. Others require advances across cryptography, systems, and hardware. We intend to work on both.
Encrypted computationConfidential computingSecure executionCryptographic isolationVerifiable computeEphemeral executionDistributed trust
BILLIONS OF MACHINES.
ONE GLOBAL COMPUTE POOL.
ZERO REQUIRED TRUST
IN WHO OWNS THEM.
COMPUTE ANYWHERE. TRUST NOTHING.
REVEAL NAUGHT.
Help build infrastructure for computation beyond trust.
Explore the technology