Marquor
Marquor provides an agent decision assurance layer that governs each autonomous AI action before it executes, checking permissions against tool, destination, and data category in real time. The platform signs tamper-evident, Ed25519-signed receipts for every decision without reading prompts or data, and runs in the customer's own VPC with AWS FTR and WAFR approvals.
- Cybersecurity
- Regulatory & Compliance Technology
- Artificial Intelligence
- AI Agents
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
As organizations deploy autonomous AI agents to take consequential actions—exporting data, calling external APIs, or initiating transactions—there is no reliable way to govern each specific action or prove afterwards that the agent acted within approved boundaries. Existing security tools focus on guardrails, sandboxes, and identity but fail to address the critical moment when an agent acts, leaving a gap in auditability and accountability for regulated industries.
Solution
Marquor fills this missing layer by providing a decision and evidence layer for autonomous agents. Agents call through the Marquor gateway, which decides whether a specific action may proceed—checking the tool, destination, and data category—and signs a tamper-evident receipt for every decision without ever reading prompts, responses, or underlying data. The system uses an issuer-custodian pattern where Marquor verifies rather than issues agent identities, keeping the customer's IdP as the source of truth. It provides live standing checks with fleet-wide revocation in seconds and fails closed with a kill-switch path. Evidence includes Ed25519-signed, hash-chained receipts aligned with EU AI Act Art. 12, verifiable offline without trusting Marquor's dashboard.
Target Audience
Primary customers are enterprises in regulated sectors—banking, fintech, healthcare, and public sector—that deploy autonomous agents and need examinable, cryptographically verifiable controls over agent actions.
Features
- Payload-free architecture: never reads prompts, responses, or data; only signed metadata reaches the control plane
- Issuer-custodian pattern: verifies agent identity without ever becoming the identity provider, so Marquor holds nothing that could mint or replay agent identities
- Live standing: re-checks credential validity on every call and revokes agents fleet-wide in seconds without redeployment, with a fail-closed kill-switch path
- Tamper-evident evidence: Ed25519-signed, hash-chained receipts viewable in an append-only ledger, mapped to EU AI Act Art. 12, DORA, ISO/IEC 27001, and NIST CSF 2.0
- Graduated response controls: detect and contain mandate deviations and hijack signals with observe, step-up, deny, and kill modes, including a human gate before irreversible actions
- GPU-accelerated decision engine built with NVIDIA Morpheus and RAPIDS, running on AWS with FTR and WAFR approvals
- Runs in the customer's own VPC with data residency and sovereign-model routing, across AWS AgentCore, GCP ext_authz, Azure, Kubernetes, or air-gapped environments