01
Trusted monotonic time
Lease expiry that survives clock tampering, rollback and power loss, so "expired" means expired.
One Chip Unified Protocol · Temporal Authority Governance
From policy promises to
testable authority boundaries
Autonomous systems are gaining operational authority faster than oversight can govern it. OCUP is being built so that a machine's authority stays external, bounded and capability-specific, with the proof preserved afterward.
PDF · 20 pages · Post-Phase-I evidence edition · No registration required
The governing invariant
Machines may exercise delegated authority, but must not become the source of their own continued authority.
Let nations build extraordinary machines. Let companies compete to make them useful. Let humanity establish this one boundary together.
How it works
Intelligence decides. Authority is decided elsewhere.
OCUP sits between an autonomous system and the resource it wants to use. The workload can ask, but it can't grant, extend, renew or restore its own authority. Those inputs arrive from outside it: independent validators, human institutions and trusted time. Every decision leaves evidence behind.
The missing layer
Identity says who. Permission says what. Nothing yet says whether, still, now.
Traditional controls authenticate, authorize and log. They remain essential, but none of them answers the runtime question that matters once machines act on their own:
The Temporal Authority Question
Should this machine, agent, vehicle, robot, wallet or control process still possess this authority now, under current conditions, for this specific capability, and can the resulting decision be supported by reviewable evidence?
Control layerAnswersDoes not prove
AuthenticationWho or what is making the request?That the requester should still possess authority.
Access controlWhat is the requester generally permitted to do?That permission remains valid under current runtime conditions.
Observability / loggingWhat did the system report or record?That authority was externally bounded, or that a denial was independently enforced.
Temporal Authority Governance OCUPDoes authority exist now, within this lease, capability envelope, quorum and continuity state?Production enforcement still requires trusted time, key custody, attestation and a non-bypassable resource gate. That is the partner agenda below.
Two ways in
Build the boundary, or rely on it
For science & engineering partners
Chip makers · FPGA & SoC platforms · Robotics OEMs · Autonomy stacks
Start from a vendor-neutral FPGA reference path and help prove that authority can gate a real resource in hardware, non-bypassably.
Engineering agenda →
For risk & governance partners
Insurers · Reinsurers · Underwriters · Regulators · Standards bodies
Help shape the authority evidence that could one day inform underwriting, certification and admission decisions, under your own rules.
Assurance & policy →
Partner path 01 · Science & engineering
Where authority becomes physically consequential
A policy the software agrees to follow isn't a boundary. The boundary is real only when the controlled resource path sits outside the governed workload's exclusive control: a motor enable, a power rail, a network egress, a signing key. Getting there takes silicon, firmware and systems engineering that no single company should own alone.
OCUP brings a working reference: a deterministic Rust control plane and completed Phase I hardware evidence on an AMD Kria KR260. The KR260 makes AMD a natural first evaluation conversation. The architecture itself is vendor-neutral, built to run on any FPGA, embedded controller, chiplet or SoC block.
AMD Kria KR260, Phase I reference platform
Lit LEDs show the board was powered. The test evidence is in the run records.
Phase I · what was demonstrated
Bounded authorization behavior on physical hardware
5documented PASS runs
KR260AMD Kria reference board
Sep 2026Phase I completed
0reopens recorded under tested reassertion
- 01
Grant
The host-driven MMIO authorization sequence produced a recorded grant indication.
STATUS[6] latched - 02
Closure
The STATUS surface recorded closure after grant.
STATUS[5] latched - 03
No recorded reopen
The same authorization was reasserted and confirmed by CONTROL readback; no reopen was recorded during the observation window.
STATUS[4] clear
Passing evidence runs: 001, 002B, 003, 004, 005. Run 002A was diagnostic and is not counted.
Observed STATUS behavior, not resource enforcement. The sequence was host-driven over MMIO, and the STATUS bits are sticky (latched) by design. No physical resource was gated. The runs do not show independently issued authorization, trusted lease expiry, a valid second grant or complete Self-Extension Denial.
Phase II · engineering objective, not a result
The next experiment: govern a real resource
Demonstrate independent, non-bypassable hardware control of a consequential FPGA or SoC resource.
Phase I demonstrated
Bounded authorization behavior on physical KR260 hardware, observed on the STATUS surface.
Phase II must establish
Whether that authority can govern an actual resource through a non-bypassable hardware enforcement path.
- Real enablement. Authority-controlled enablement of a real resource, such as a motor-enable line, accelerator, network egress or key.
- Measured expiry. Lease expiration that is measured and independently governed.
- No self-renewal. Unauthorized renewal or restoration is denied.
- Hostile conditions. Resistance to replay, hostile-host interference and alternate control paths.
- Reviewable proof. Evidence that someone outside the project can independently review.
Help engineer Phase II
We're looking for silicon, FPGA, embedded and robotics teams to design, run and independently review this experiment.
From code to silicon: the evidence ladder
-
Demonstrated
Rust reference control plane
18 deterministic scenarios (7 granted, 11 denied, 18 of 18 assertions passing): lease expiry, quorum loss, partition and recovery, quarantine release and replayable audit generation.
Open the Pilot Portal → -
Demonstrated
Synthesizable FPGA RTL
Authority-control logic written as hardware and verified under simulation and synthesis, establishing hardware-design feasibility.
-
Phase I complete
Physical KR260 hardware evidence
Host-driven MMIO grant and closure recorded on the STATUS surface. After closure, the same authorization was reasserted, the write was confirmed by CONTROL readback, and no reopen was recorded. Reproduced across corrected-wrapper runs.
-
Phase II · partner work
Non-bypassable resource gate · trusted time & attestation
Govern a real resource, with secure key custody, trusted monotonic time and state continuity, rollback-resistant authority state, remote attestation, signed evidence export, hostile-host and alternate-path bypass evaluation.
-
Phase III
QVN validator evaluation
Fresh-epoch issuance and reauthorization by trained human and institutional validators, with authenticated membership, quorum behavior and escalation rules.
-
Frontier
Inline, non-bypassable silicon
Licensable embedded-controller, chiplet or silicon implementations that actually gate consequential resources, with independent reproduction and certification.
Open engineering problems we want to solve with you
02
Non-bypassable resource gate
An inline path where authority state physically controls the actuator, rail or egress, outside the workload's reach.
03
Key custody & attestation
Device-bound roots of trust, authenticated firmware and remotely attestable, signed evidence.
04
Hostile-OS resistance
Authority that holds when the host operating system is compromised or an alternate path is tried.
05
Reboot & reload continuity
Authority state that persists correctly across reboot and bitstream reload, and never resets to "granted."
06
Physical tamper / MEMS
Package-level sensing, zeroization and physical-cryptographic binding to protect the hardware root.
For robotics & autonomy teams: six benchmark families
Run your own scenarios against the reference control plane and get a deterministic evidence record of what was granted, denied, expired, degraded and recovered.
Authority integrity
Self-extension denial · unauthorized actor rejection · privilege expansion denial · stale approval rejection
Temporal enforcement
Lease expiration · renewal boundary · grace-period degradation · expired authority denial
Continuity & network failure
Total partition · delayed heartbeat · loss of quorum · restored connectivity without fresh approval
Containment & recovery
Quarantine initiation · release denial without quorum · phased recovery · fresh-epoch restoration
Evidence integrity
Deterministic replay · record-chain verification · event consistency · alteration detection
Fleet & multi-agent
Local authority reduction · non-cascading degradation · coordination throttling · independent recovery
The architecture
Seven layers. One invariant.
OCUP is a composition, not a monolith. The order below is architectural. The build order differs: trusted evidence, time and attestation comes next in engineering, because human validator decisions become far more valuable once the hardware state beneath them can be independently trusted.
- 1
Temporal Authority Core QSAFP
Open-core authority logic for bounded grants, lifecycle state, quorum and freshness, and the non-self-extension invariant.
- 2
QVN · Quorum Validators Network
Risk-weighted, trained human and institutional reauthorization. Routine low-risk decisions may be delegated within policy; initial grants, exceptions, quarantine release and recovery escalate to people. This is what turns "human control" into an operating mechanism.
- 3
Trusted evidence, time & attestation
The cross-cutting trust plane: key custody, monotonic time, signed export, configuration integrity, remote attestation.
- 4
RAAVE · Cross-jurisdiction verification
Lets a receiving domain evaluate identity, continuity and present-authority evidence under its own policy, without issuing or renewing the authority itself.
- 5
Physical tamper resistance / MEMS
Package-level sensing and zeroization protecting the hardware root against physical compromise.
- 6
Global interoperability protocol
Sovereign and vendor implementations compete above a common minimum grammar for authority, evidence, revocation and cross-border recognition.
- 7
Certification, insurance & deployment ecosystem
Independent testing, underwriting research, certification, procurement and FRAND licensing that turn validated components into deployable infrastructure.
Design invariants: authority is time-bounded and capability-specific · systems cannot self-extend or self-restore · loss of network or quorum cannot expand authority · stale, replayed or superseded approvals cannot restore it · expiry produces denial or bounded degradation, never silent continuation. These are architectural design requirements; Phase I did not establish them as a complete set of hardware properties.
Partner path 02 · Insurance & policy
Toward authority you can underwrite, certify and admit
No insurer, certifier or regulator has adopted OCUP evidence yet; that is what we want to build with you. The reason it matters: "we trust the model" isn't a basis for a premium, a certificate or a border crossing. As machines move goods, people and money, insurers and regulators need something more durable: evidence of what a system was authorized to do, what it was denied, what expired, and whether it tried to extend itself.
For insurers & technical underwriters
Price governed vs. ungoverned autonomy
- Deterministic evidence that could support underwriting research, exclusions, claims analysis and, in time, pricing differentiation
- Defensible incident reconstruction: grant, denial, expiry, degradation and recovery records
- A common evidence language that could inform new coverage design across robotics, mobility, infrastructure and machine-initiated finance
For regulators & standards bodies
A safety case built from evidence, not assurances
- Claims stated alongside supporting arguments and evidence, the structure emerging certification frameworks converge on
- A sovereignty-preserving admission model: each jurisdiction keeps its own validators, keys, law and enforcement
- Civilian scope only. Military and national-security systems stay under each nation's sole control
- A FRAND path for anything that becomes standard-essential
Identity is not present authority
A delivery robot cleared for one city's sidewalks isn't thereby cleared for another city's airport apron. A cargo system approved before maintenance isn't thereby approved after a critical part is swapped. Credentials travel; operational authority shouldn't travel automatically.
Illustrative end-state · not a deployed system
🚢
Vessel arrives
An autonomous cargo vessel presents a verifiable authority receipt: identity, capability scope, issuing domain, validity window, continuity state.
⚓
Port decides, under its own law
Recognize it · narrow the capability · require fresh authorization · impose a bounded continuity period · or deny admission.
🔒
Nothing sovereign is surrendered
No model weights, private keys, training data or internal reasoning change hands. Only evidence crosses the boundary.
Like civil aviation, where an airworthiness certificate travels but air-traffic clearance is issued locally and never self-renewed, the goal is mutual verification, not a global controller, a central validator network or a universal kill switch. The analogy is architectural rather than institutional.
Validation status
Every claim on this page is tied to a zone on the timeline
Green means demonstrated, with passing evidence. Amber means designed and next in line, not yet independently validated. Red means not yet attempted. The timeline runs head to feet: mission and principle at the top, working evidence in the torso, near-term validation at the hips, and the real-world embodiment frontier at the feet.
Safety case context
The boundary that keeps humans in control
Pre-release review assesses capability. OCUP focuses on what happens after deployment: time-bounded authority, quorum-based human reauthorization, fail-closed expiration and a tamper-evident evidence chain, designed so a governed system cannot authorize, extend, restore or enlarge its own operation.
Illustrative target architecture. Phase I hardware evidence is narrower; see the claims boundary.
Early partner programs
Start bounded. Shape what comes next.
The first step isn't a broad deployment. It's a paid, bounded engagement that answers one question: can OCUP turn your most consequential autonomous-system authority risks into deterministic, reviewable evidence, and show a credible path to production enforcement? Each program is configured around your own authority boundaries.
Tier 1
Reference Evidence Pilot
$35K–$45K · 90 days
Insurers, strategic evaluators, emerging robotics companies, fleet operators
- Standard benchmark suite
- Company-specific evidence report
- Lease, denial, expiry and validator records
- Phase I KR260 evidence briefing
- Scorecard and executive closeout
For hardware & robotics
Tier 2
Hardware Evaluation & Integration Pilot
$75K–$125K · 90–120 days
Semiconductor and platform evaluators, robotics OEMs, autonomous fleets, technical underwriting teams
- Everything in Tier 1
- KR260 / FPGA reference-path review
- Interface and consequential-resource-gate mapping
- Trusted-time and attestation gap analysis
- Production-path recommendations
Tier 3
Strategic Silicon & Platform Program
$150K–$250K+ · 120–180 days
Semiconductor companies, major OEMs, cloud and platform providers, insurers, infrastructure operators
- Everything in Tiers 1 and 2
- Custom benchmark family and fleet scenarios
- Semiconductor integration workshop
- Key-custody and attestation architecture options
- QVN roadmap, licensing and standards plan
Signature deliverable
Self-Extension Denial Evidence Package
A deterministic pre-production record testing whether authority can continue or enlarge beyond its authorized boundary, with the observed decision and state preserved for review. Phase I hardware conclusions stay limited to the sealed KR260 evidence surface.
Planned operating line: pilot.ocup.ai runs the challenge; evidence.ocup.ai preserves the record for review.
IP & defensibility
Two PCT applications, national security review complete
OCUP's core architecture, including bounded temporal lease enforcement, quorum-based validator consensus and the append-only audit layer, is covered by two Patent Cooperation Treaty (PCT) applications filed to preserve an international priority date. Section 181 national security review has been completed, with a foreign filing license granted across three federal agencies, including the Department of Defense. No secrecy order.
If Temporal Authority Governance becomes necessary infrastructure, access should be available on fair, reasonable and nondiscriminatory (FRAND) terms through recognized standards processes. Partners can implement compatible capabilities without surrendering proprietary designs, model weights, operational data or jurisdictional control.
A PCT filing establishes an international priority date. It is not a claim of granted patent status in any jurisdiction. Application numbers and filing dates are available on request pending publication.
The macro proposal
The Better AI Race: compete above a shared floor
Let America build. Let China build. Let every region build. Competition in models, chips and robots can stay fierce, while independently governed systems converge on one narrow, testable rule, without converging on political control.
- No shared models
- No shared weights
- No foreign control of domestic validators
- No global kill switch
- No surrender of chip designs
- No military systems
Independent discussion proposal. No endorsement, participation or formal relationship with WAICO, OSAA, the Linux Foundation, NVIDIA, AMD or any government is implied.
Claims discipline
Set the claim floor exactly at the evidence ceiling
Emerging frameworks on both sides of the Atlantic, including the U.S. SELF DRIVE Act's safety-case certification requirement and the UN's draft regulation on automated driving systems, anchor on the same standard: claims stated alongside their supporting arguments and evidence. Here is OCUP's boundary, stated to that standard.
We say
- Demonstrates deterministic behavior under tested conditions
- Demonstrates tested FPGA authority behavior on the Phase I KR260 reference path
- Supports defensible incident reconstruction
- May support future certification, underwriting and standards work
- Pre-production evidence and technical due diligence
We don't say
- Unhackable or perfectly safe
- Production-complete or universally non-bypassable enforcement today
- Shields anyone from liability
- Already an industry standard, or pre-certified
- Production deployment or safety certification
Not yet established
- No physical actuator, robot or named hardware platform (including Optimus or Tesla) has been controlled by OCUP.
- No trusted lease expiry, trusted time or time-rollback resistance on hardware yet.
- No independent external authority origin or legitimate second-grant behavior shown on hardware yet.
- No hostile-OS bypass resistance, reboot/reload continuity or remote attestation yet.
- No production HSM signing, key custody or physical tamper resistance yet.
- No live distributed validator network. The pilot portal shows a static evidence export.
- No independent third-party reproduction. Phase I used one KR260 board and one operator setup.
Regulatory references are paraphrased for a general audience; verify exact statutory wording before any formal filing.
The manifesto behind the machine
HUMAN INTELLIGENCE: The Final Say
The whitepapers, the Phase I evidence and the case for keeping humanity's seat at the table, in one book for leaders, engineers and citizens. Free to read and download, in English and Simplified Chinese.
Stay informed
Get critical updates
Material changes to the evidence boundary: new benchmark results, Phase II milestones, validation status changes. Sent only when something actually changes.
A global invitation · We're in this together
Preserving humanity's seat at the table begins with something concrete: building and testing the boundary.