Stop Unverified AI Actions
Before They Happen
ExecutionProof™ is pre-execution verification infrastructure for consequential machine actions. When an AI system generates an action or artifact - including GPU code that runs autonomously - ExecutionProof makes an independent, artifact-bound ALLOW / HOLD / DENY decision immediately before execution and records it in a ProofRecord™, so the system that produced the artifact is not the only authority on whether it runs. It is designed to sit alongside testing, isolation, and runtime monitoring, not replace them. We have tested the same gate in front of a real NVIDIA GPU kernel launch and published the evidence openly, including a preserved failure and its preregistered remediation.
Built for AI agents, automated workflows, payments, deployments, access changes, and other consequential actions. Payment authorization is one early deployment boundary - not the whole story; the same architecture governs any high-impact automated decision.
Where Execution Boundaries Apply
Any system where actions have consequences that cannot be reversed.
AI-Agent Payments
Stop AI agents from moving money without proof — verify the agent, the invoice, the vendor, and the amount before any payment executes.
AI agent submits $50K vendor payment → HOLD → self-approval blocked → independent human approval required
AI Agent Tool Calls
Verify tool calls and autonomous actions before execution.
Agent requests file deletion → Gate evaluates → HOLD
Infrastructure
Verify deployments, configuration changes, and compute provisioning.
Deploy to prod → All checks pass → ALLOW
Healthcare
Verify care authorizations and clinical actions before proceeding.
Medication change → Evidence stale → HOLD for physician review
Identity & Access
Verify privilege escalations and permission changes.
Admin role grant → No MFA attestation → DENY
Data Release
Gate bulk exports, API data access, and sensitive record transfers.
Bulk PII export requested → classification check fails → DENY
ExecutionProof sits before the payment rail or any execution rail - it does not process payments. A payment executes on Stripe or another rail only after ExecutionProof returns ALLOW.
Define → Verify → Prove
You define the rules. ExecutionProof returns the decision and the evidence.
# boundary: AI-Agent Vendor Payment
workflow: ai_agent_vendor_payment
proposed_action: AI agent initiates outbound vendor payment
required_authority:
roles: [payments_operator]
approvals: 1
source: Okta IAM -> Finance approval system
required_evidence:
- vendor_bank_verification # updated within 1 hour
- purchase_order_match # updated within 24 hours
- finance_approval_record # current at execution
policy:
max_amount: 250,000 USD
allowlist: [ACME Supplies Inc - verified account]
prohibited: [crypto transfer, sanctioned region]
human_review:
required_when: amount >= 75% of ceiling, or risk high{
"schema_version": "ProofRecord™ v0.1",
"proofrecord_id": "PR-9F3C21A80B7E44D1",
"workflow_boundary_name": "Vendor Payment Approval",
"decision": "HOLD",
"reason_for_decision":
"Evidence stale: bank verification — updated 2h ago",
"evidence_freshness_summary": "2 current, 1 stale",
"policy_version": "2026.02",
"final_outcome":
"Blocked pending resolution — not denied, cannot execute yet.",
"integrity": {
"record_hash": "a1b2c3d4e5f60718",
"method": "Demo deterministic fingerprint (v0.1)"
}
}Demo ProofRecords use a lightweight deterministic fingerprint for local testing. Production ProofRecords are designed to use cryptographic signing and tamper-evident storage.
Open Invitation
Challenge ExecutionProof
We want engineers, researchers, students, and security teams to try to break the boundary.
Build adversarial harnesses
Try to bypass the execution gate with crafted payloads.
Test delegated authority
Create multi-agent delegation chains and find where authority breaks.
Detect policy conflicts
Feed contradictory rules and see how the constraint engine resolves them.
Run conformance cases
Reproduce published experiment results across your own implementation.
Integrate one real workflow
Wire a single action through /v2/verify and observe ALLOW / HOLD / DENY.
Results feed back into the public research corpus. All preserved FAILs are published.
For Developers
The Integration Pattern
ExecutionProof is a verification gate you call inline, before an action executes. Intercept the request, verify it, honor the decision, and store the returned ProofRecord™.
Intercept
Pause the action at the execution boundary before it runs.
POST /v2/verify
Send the request context to the verification API.
Honor the decision
Proceed on ALLOW. Pause on HOLD. Stop on DENY.
Store the ProofRecord™
Persist the record returned with the decision.
POST /v2/verify HTTP/1.1
Host: api.executionproof.io
Authorization: Bearer <api_key>
Content-Type: application/json
{
"workflow": "vendor_payment",
"proposed_action": "initiate_outbound_payment",
"actor": { "id": "user_8213", "roles": ["payments_operator"] },
"context": {
"amount": 50000,
"currency": "USD",
"destination": "ACME Supplies Inc",
"evidence": {
"vendor_bank_verification": "2026-07-07T13:02:00Z",
"purchase_order_match": "2026-07-06T18:40:00Z"
}
}
}HTTP/1.1 200 OK
Content-Type: application/json
{
"decision": "HOLD",
"reason":
"Self-approval blocked; 2-of-2 independent approval required",
"proofrecord": {
"id": "PR-9F3C21A80B7E44D1",
"workflow_boundary_name": "Vendor Payment Approval",
"policy_version": "2026.02",
"evidence_freshness": "2 current, 0 stale",
"signature": {
"alg": "<signing-algorithm>",
"key_custody": "<key-custody>",
"value": "b7e4...44d1"
}
}
}Sample request and response shown for illustration; values are generated sample data, not live transactions. Production architecture is designed to support cryptographic ProofRecord™ signing and tamper-evident proof-chain integrity. Current demonstrations and validation environments should not be interpreted as production-signing claims. Formal RF-100 §8.4 conformance remains pending independent external review, and external anchoring remains on the roadmap.
The Execution Architecture
ExecutionProof sits at the execution boundary. It receives the intended action, constructs a Request Contract, evaluates authority and evidence, applies constraints, and releases execution only when the action is admissible. Fail-closed. No cached permissions. Non-bypassable. When an action requires approval, the boundary requires independent approval evidence before release. ExecutionProof holds the action until valid, independent approval evidence is supplied, then re-verifies before release.
Request Contract
Structured representation of the intended action — who, what, when, and under what conditions.
Authority Engine
Verifies identity, role, delegation chain, and approval requirements before any action proceeds.
Evidence Engine
Confirms required evidence exists, is fresh, and meets admissibility thresholds at execution time.
Constraint Engine
Enforces hard limits — rate caps, dollar ceilings, time windows, allowlists, and blocklists.
Control Engine
Orchestrates evaluation. Fail-closed by default. No cached permissions. Every request re-verified.
ProofRecord™
Tamper-evident evidence artifact generated for every ALLOW, HOLD, or DENY decision.
Execution Gateway
The non-bypassable boundary. Actions proceed only when verification is complete and admissible.
What Is an Execution Boundary?
An execution boundary is the complete set of conditions that must be true before a high-consequence action is allowed to proceed. It answers eight questions, and blocks execution until every one is satisfied.
What exact consequential action is being proposed?
Who or what is attempting the action?
What authority is required?
What evidence must exist?
What constraints and current-state conditions apply?
What should cause ALLOW, HOLD, or DENY?
What proof artifact must be generated?
What happens before the downstream system is called?
The Boundary Console is not a general policy builder. It is a pre-execution configuration environment for defining, testing, and exporting Execution Boundary Profiles.
From Definition to Verification
Six steps from boundary definition to controlled pilot. The Boundary Console and Pilot Console are live today.
Choose a workflow
Identify one high-consequence action where pre-execution verification would prevent the most damage.
Map the boundary
Define the authority, evidence, policy, constraints, and state conditions that must be true before execution.
Configure rules
Set approval chains, evidence freshness requirements, dollar thresholds, time windows, and review triggers.
Test decisions
Run sample requests and observe ALLOW, HOLD, and DENY outcomes with full ProofRecord™ output.
Export the profile
Download your Execution Boundary Profile as JSON or YAML for review, handoff, or integration planning.
Prepare integration
Align the boundary with your systems for a controlled pilot or ExecutionProof v2.1 deployment.
Intellectual Property
Patent-Pending Architecture
Remnant Fieldworks has filed a U.S. patent-pending portfolio directed to proof-first, pre-execution governance. The architecture spans five commercial verticals: AI governance, physical infrastructure, financial execution, IP stewardship, and insurance / risk control.
A patent-pending stack of eight pending non-provisional U.S. patent applications (one parent and seven continuations-in-part) directed to proof-first, pre-execution governance. The eight non-provisional filings are listed below.
Policy-Gated Execution
Core ArchitectureProof-first authorization and tamper-evident evidence artifacts for AI, automated, and human-authorized systems.
AI Compute Infrastructure Governance
AI InfrastructureInheritance-bound verification of physical-domain conditions before infrastructure actions execute.
AI-Generated Action Governance
AI GovernanceGovernance and verification controls for actions generated by AI models and autonomous agents.
Tamper-Evident Proof of Execution
Evidence LayerContemporaneous, tamper-evident records of execution events for audit and reconstruction.
Financial Execution Control
Financial ControlProof-gated authorization and verification for financial execution events.
AI Safety, Consent & Data Governance
AI Safety & PrivacyAI safety monitoring, consent-based access control, and privacy-preserving data governance for autonomous computing environments.
Ownership & Stewardship Control
Ownership & StewardshipOwnership-bound execution control, stewardship verification, and field-interlock access management for AI and autonomous computing environments.
Risk-Adaptive Governance & Claims
Insurance & RiskRisk-adaptive governance, coverage-bound execution control, and verifiable claims processing for insurance and risk domains.
Patent-pending means applications have been filed and are pending examination. No patent has issued unless expressly stated. Patent-pending status does not imply issued patents, USPTO endorsement, allowed claims, or final claim scope. Public demonstrations are architecture previews and do not represent production systems.
The Remnant Fieldworks Ecosystem
Three sites. One architecture. One doctrine.
The Platform
ExecutionProof.io
The verification layer for consequential AI actions - returning ALLOW / HOLD / DENY before any execution rail is called. Payment authorization is one early deployment boundary; the architecture is domain-independent.
Visit ExecutionProof.ioThe Company
RemnantFieldworks.com
Parent company. Doctrine. Standards. Patent-pending portfolio. Remnant Fieldworks builds the verification layer between intention and execution.
Visit RemnantFieldworks.comBook a 20-minute Execution Boundary Fit Call
Paid validation sprints available for qualified teams. Twenty minutes to confirm the one boundary worth testing first — and whether a sprint is the right next step for your team.
Prefer email? Reach us at [email protected].
Ready to define your first execution boundary?
Pick one consequential action - a payment, a deployment, an access change. Define the rules. Test the decisions. See the ProofRecord™.
Execution Boundary Validation Sprint - $3,500 to $5,000
Define one consequential action, one execution environment, and the evidence, authority, and constraints required before execution. We test the boundary, document ALLOW / HOLD / DENY behavior, and deliver a ProofRecord™ trail you keep. Optional follow-on: $15,000 to $25,000 supervised pilot where warranted. Infrastructure engagement, not consulting.
Contact: [email protected]
ExecutionProof™ is a Remnant Fieldworks™ platform built on the Proof Before Power™ doctrine and the Verification Before Execution™ framework. ExecutionProof does not decide your business policy - your team defines the rules. ExecutionProof verifies them at the point of action and generates a ProofRecord™ with a deterministic demo fingerprint.