Intent can drift into effect.
Agent reasoning, tool permissions and execution paths can become coupled in ways that make duplicate, stale or insufficiently authorized actions harder to contain and audit.
MODEL→TOOL→EFFECT
VSL-ECP · External Effect Control Plane
AI can decide.
Authorization still matters.
Authorization infrastructure for autonomous AI actions. VSL-ECP separates model reasoning from permission to create consequential external effects.
NO VALID PERMIT = NO EXTERNAL EFFECT
PRODUCT DIRECTION / DESIGN PRINCIPLENO EXTERNAL EFFECT WITHOUT COMPLETE, VERIFIABLE EFFECT PROVENANCE.
PHYSICAL CONCEPT / VSL-ECP
15 SEC · AUTOPLAY LOOP
A physical visualization of the control boundary. The video loops automatically; sound is optional. The validated prototype is software infrastructure.
20independent Cloud Run tasks
1successful consumption
19already consumed
0exceptions
The control gap
Reasoning is not authorization.
Autonomous systems are moving from generating answers to using tools, changing infrastructure, executing workflows and affecting real systems. A model can propose an action without the enterprise having authorized that exact consequence.
VSL-ECP is designed for the boundary between reasoning → authorization → execution → evidence.
An effect crosses a separate authorization boundary.
The agent may request. ECP verifies whether that specific effect has valid authority, consumes it once and records the outcome.
REQUEST→PERMIT→EFFECT
Public architecture
One visible control path.
Implementation stays confidential.
The public model is intentionally simple. It explains the boundary without disclosing cryptographic internals, operational controls, credentials or private deployment details.
01AI Agent
/ Workflowproposes an action
/ Workflowproposes an action
Action request
02Valid Permitbound authority
verification
03VSL-ECPverify · state · consume · evidence
authorized
04External
Effectbounded execution
Effectbounded execution
receipt
05Auditable
Receiptrecorded outcome
Receiptrecorded outcome
Verify permit
Confirm that authority exists for the requested effect before execution is allowed.
Check state
Evaluate the transactional state of the permit at the authorization boundary.
Consume once
A successfully redeemed permit becomes unavailable to competing consumers in the tested model.
Record evidence
Preserve an auditable record of the authorization and observed outcome.
Validated behavior
A bounded Google Cloud prototype survived distributed contention.
In the recorded test, 20 independent Cloud Run tasks contended for the same permit. One task consumed it successfully. The other 19 observed that the permit had already been consumed. Cloud Spanner persisted the final consumed state and the winning task.
VALIDATED TECHNICAL PROTOTYPE · NOT PRODUCTION CERTIFICATION
G1.3C-001PASS
CONTENDERS20
SUCCESS1
ALREADY_CONSUMED19
EXCEPTIONS0
PERSISTED STATECONSUMED
VERSION2
Observed single-consumption behavior under the tested contention conditions.
Validated bounded Google Cloud experiment
- 20 independent Cloud Run tasks
- 1 SUCCESS
- 19 ALREADY_CONSUMED
- 0 exceptions
- Cloud Spanner persisted final consumed state and winner
Validated technical prototype
Authorization-to-effect provenance
- authorization receipt and effect receipt separated
- CONFIRMED / FAILED / UNKNOWN effect semantics
- actor / workload / software-policy version provenance
- causal trace and incident reconstruction support
- tamper-evidence design and configurable regulatory profiles
LOCAL_VALIDATED
Cloud integration validation pending for this extension.
Authorization → observed outcome
From authorization to effect provenance.
A valid permit proves authority to attempt an effect. It does not prove that the external effect actually occurred.
VSL-ECP is evolving from a permit-control boundary into a verifiable effect-provenance layer that can reconstruct what was intended, what was authorized, what was attempted, what was actually observed and under which software/policy version the action occurred.
INTENT→PROPOSED ACTION→AUTHORIZATION→PERMIT→EXECUTION ATTEMPT→OBSERVED EFFECT→RECEIPT→POST-EFFECT STATE
Evidence indicates that the intended external effect occurred.
The attempt did not produce the intended effect.
Available evidence is insufficient to assert either success or failure.
VALID PERMIT + FAILED EFFECT ≠ SUCCESS
VALID PERMIT + UNKNOWN EFFECT ≠ SUCCESS
NO RECEIPT ≠ PROOF OF NO EFFECT
Provenance extension: LOCAL_VALIDATED.
Cloud integration validation pending.
Forensic surface
Reconstruct the chain, not just the log.
For each bounded effect trace, the provenance architecture is designed to preserve verifiable references without publishing confidential schemas or operational identifiers.
Who proposed it. Who authorized it. What executed it. What actually happened. Under which version.
agent intentproposed actionauthorization requestpermit decisionpermit validationexecution attemptobserved external effectauthorization receipteffect receiptpost-effect stateactor and workload identitiessoftware / policy / schema versionincident and correction lifecycle when applicable
Interactive protocol demonstration
See the authorization boundary behave.
This browser-only demonstration contains no production credential, confidential implementation or consequential external effect. It illustrates the public state model and the separation between permit consumption and observed effect outcome.
Interactive protocol demonstration — not a production system.
AGENTAction requestWaiting
PERMITNoneNo authority loaded
ECPStandbyauthorization boundary
EFFECTBlockedexternal side
RECEIPT—auditable evidence
DEMO AUTHORIZATION RECEIPT
SIMULATED PROVIDER OUTCOMEAuthorization is not effect confirmation.
Select an observed outcome only after a permit has been consumed.
PUBLIC-SAFE EFFECT RECEIPT
Regulatory context
Incident reconstruction is becoming operationally consequential.
From 11 September 2026, the EU Cyber Resilience Act begins applying its Article 14 reporting obligations for actively exploited vulnerabilities and severe security incidents affecting products with digital elements. These obligations increase the operational value of being able to reconstruct when an incident was detected, which software version was affected, what mitigation or correction followed, and what actually occurred.
VSL-ECP’s provenance roadmap is designed to support this kind of verifiable authorization, effect and incident timeline reconstruction. It does not itself establish legal compliance, certification or applicability under the CRA; those determinations depend on the organization, product, jurisdiction and legal assessment.
Evidence support for incident reconstruction — not automatic legal compliance.
European Commission · CRA reporting obligations ↗
Enterprise fit
Control consequential AI actions before they become external effects.
We are seeking a limited number of qualified enterprise design partners to test bounded authorization and effect-provenance workflows in real integration contexts. ECP does not replace identity, policy, compliance systems or human accountability.
Production changes
Gate deployment or configuration effects behind explicit effect authority.
Agent tool execution
Separate model reasoning from the authority to invoke consequential tools.
Enterprise workflows
Add a consumption and evidence boundary to high-impact automated steps.
Procurement & approvals
Bind delegated approval to the specific effect that may occur.
Consequential operations
Reduce ambiguity between requested, authorized and observed execution.
Evidence paths
Leave a clear record of what authority was consumed and what outcome was observed.
PUBLIC CONCEPT · CONFIDENTIAL IMPLEMENTATION
Enough to evaluate.
Not enough to expose the mechanism.
The public VSL-ECP page explains the product boundary, validated behavior and high-level architecture. Detailed technical architecture, validation artifacts and implementation materials are shared selectively with qualified prospective investors and approved design partners under controlled access.
The public site intentionally does not expose source implementation, credentials, key material, cryptographic internals, unpublished threat-model details, internal deployment identifiers or partner-specific integration design.
Investor / Design-Partner Diligence
For qualified investors, enterprise design partners and technical diligence.
Request controlled diligence access
- Problem definition
- Control-path model
- Bounded cloud validation result
- LOCAL_VALIDATED provenance behavior
- Enterprise use cases
- Source implementation
- Cryptographic internals
- Credentials and key material
- Unpublished threat-model details
- Partner-specific integration design
Vehicle Systems Lab ecosystem
A focused enterprise product direction inside a broader research laboratory.
VSL-ECP is being packaged as a focused enterprise control-plane product. The broader Vehicle Systems Lab ecosystem supplies research context and technical lineage, but the public ECP surface stays centered on authorization, execution and evidence.
Research lineage: CRLP explores pre-production commit authorization and external-effect containment. VSL-ECP is the focused enterprise product direction for authorization-to-effect control and evidence.
Validated prototype environment
Built and tested on Google Cloud infrastructure.
The current bounded prototype uses Google Cloud services including Cloud Run and Cloud Spanner to exercise distributed contention and persist permit state.
Startup infrastructure benefit: Vehicle Systems Lab has received startup cloud credits through the Google for Startups Cloud Program.
Official Google for Startups Cloud Program ↗CLAIM BOUNDARY
Use of Google Cloud infrastructure does not imply Google endorsement, certification, funding, adoption or approval of VSL-ECP.
Enterprise design partnership
Control consequential AI actions before they become external effects.
We are seeking a limited number of qualified enterprise design partners to test bounded authorization and effect-provenance workflows in real integration contexts.
For investors
From validated authorization to pilotable enterprise infrastructure.
VSL-ECP is the lead commercial product of Vehicle Systems Lab. The current opportunity is to convert a validated authorization property and locally validated provenance architecture into pilotable enterprise infrastructure.
Detailed investor and technical diligence materials are available under controlled access.
Request investor materialsClaim boundary
Ambitious, but not exaggerated.
What this public page supports
Two distinct evidence tiers: a validated bounded Google Cloud single-consumption experiment, and a provenance extension classified LOCAL_VALIDATED in controlled local testing. Cloud integration validation for the provenance extension remains pending.
What this page does not claim
No production certification, security certification, completed compliance, universal exactly-once external effects, paying customers, revenue, or replacement of IAM, policy engines, compliance systems or human review.
COMPANY MEMBERSHIP · STARTUP ECOSYSTEM
NVIDIA Inception Program
Vehicle Systems Lab is a member of NVIDIA Inception.
NVIDIA describes Inception as a free program that guides AI startups through the NVIDIA platform and ecosystem. Publicly listed member benefits include developer tools and training, preferred pricing on select NVIDIA hardware and software, exclusive partner offers and exposure to a global investor ecosystem.
Official NVIDIA Inception page ↗CLAIM BOUNDARY
Membership applies to Vehicle Systems Lab as a company. It does not imply NVIDIA endorsement, funding, certification, sponsorship, technical validation, adoption or approval of VSL-ECP.
NVIDIA and the NVIDIA logo are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of their respective owners.
VSL-ECP
One valid permit.
One authorized effect.
One auditable consumption record.
Authorization infrastructure for autonomous AI actions.