Shared Responsibility Matrix

Document Reference: AUC-SEC-SRM-001
Version: 3.2.0
Effective Date: October 24, 2026
Last Updated: October 24, 2026
Classification: Public / Customer Guidance
Standards Frameworks: SOC 2 Control Alignment | ISO/IEC 27001 Alignment | ISO/IEC 42001 Alignment | NIST AI RMF | FedRAMP / CMMC Moderate Baseline Alignment
Companion Documents: Master Services Agreement | Schedule 1: AI Service-Specific Terms | Customer Copyright & IP Protection Addendum | Sub-processor Schedule

Executive Summary (ISO 24495-1 Plain-Language Overview)

[!IMPORTANT] Legal Precedence Notice: This plain-language summary is provided solely for reader convenience and architectural navigation pursuant to ISO 24495-1:2023. It reflects operational guidance and tripartite division of responsibility. In the event of any conflict, governing contractual terms and operative addenda strictly control.

• Tripartite Model: Security and compliance are co-managed across three tiers: (1) Upstream Cloud & Model Providers (physical facilities, virtualization, base weights); (2) Auctoric (multi-model routing, runtime guardrails, tenant isolation, CCC defense); and (3) Customer (identity, data classification, human oversight).
• AI Safety & Guardrails: Auctoric provides edge PII masking and prompt injection classifiers; customers must keep guardrails active and configure custom domain blocklists.
• OWASP LLM Mitigations: Delineates technical responsibilities across prompt injection (LLM01), sensitive disclosure (LLM02), and excessive agency (LLM06).
• Enterprise RACI Matrix: Provides granular Responsible, Accountable, Consulted, and Informed mapping across infrastructure, SOC 2, EU AI Act, HIPAA, and DSAR tasks.

High-Intent Shared Responsibility FAQ (AEO Direct Answers)

Who is responsible for preventing prompt injection and adversarial jailbreaking?

Under Section 3.1, responsibility is shared. Auctoric provides real-time neural and heuristic prompt injection classifiers and context isolation; customers are responsible for parameter validation, escaping untrusted variables, and provisioning least-privilege agent permissions.

Who is responsible for human oversight (HITL) in High-Risk AI deployments?

Under Section 2 and Section 4, the Customer (Deployer) holds primary accountability. Auctoric provides technical review queues, attribution telemetry, and intervention APIs, but the customer must staff qualified personnel to review and execute final decisions.

How does Auctoric allocate responsibility for output copyright infringement claims?

Under Section 2 (incorporating the CCC Addendum), Auctoric provides primary defense and indemnification for eligible outputs, provided the customer fulfills the condition precedent of keeping platform guardrails enabled and avoiding intentional bad-faith infringement prompts.

1. Overview and Architecture Principles

Security, intellectual property protection, and regulatory compliance within the Auctoric Governed AI Platform operate under a Tripartite Shared Responsibility Model. Ensuring end-to-end security, data protection, and statutory alignment is a collaborative obligation shared among:

1. Cloud Service & Foundation Model Providers (CSP): Underlying hyperscaler infrastructure (as enumerated in our Sub-processor Schedule) providing physical facilities, specialized GPU hardware (TPUs/GPUs), foundational pre-training datasets, and base model weights.
2. Auctoric (AI Platform & Governance Provider): Multi-model routing gateway, real-time safety guardrails, PII/PCI redaction, Customer Copyright Commitment defense, agentic orchestration, tenant vector isolation, and tamper-resistant audit logging.
3. Customer (Deployer & Enterprise Tenant): User identity provisioning, prompt crafting, application data classification, client secrets management, human oversight (HITL), and downstream regulatory compliance.

flowchart TD
    subgraph CSP["1. Upstream Providers (Cloud Infrastructure & Models)"]
        Phys["Physical Data Center Facilities & Server Hardware"]
        Virt["Hypervisor, Virtualization & GPU Accelerators"]
        BaseModels["Base Foundational Model Weights & Pre-Training Lineage"]
    end

    subgraph Auctoric["2. Auctoric Platform (Governed AI-as-a-System)"]
        Kubernetes["Container Orchestration & Hardened Shielded Clusters"]
        Router["Multi-Model Gateway & Intelligent Dispatch Engine"]
        Guardrails["Runtime Guardrails, PII Masking & Safety Classifiers"]
        CCC["Customer Copyright Commitment (CCC) Output Defense"]
        AuditLog["Tamper-Resistant Audit Logging & Evidence Pipelines"]
    end

    subgraph Customer["3. Customer (Deployer & Tenant)"]
        IAM["User Account Creation, IdP Integration (SAML/SSO) & MFA"]
        DataGov["Data Classification, Sensitive Data Identification & Input Sanitation"]
        SafetyGov["Guardrail Activation, System Prompt Design & HITL Workflows"]
        AntiDistill["Adherence to Anti-Distillation & Prohibited Use Covenants"]
        Downstream["Downstream Regulatory Compliance (EU AI Act, HIPAA, CJIS)"]
    end

    CSP --- Auctoric --- Customer

2. Comprehensive Shared Responsibility Matrix

3. Detailed AI & Large Language Model (LLM) Risk Responsibilities

The OWASP Top 10 for Large Language Applications requires specific shared actions:

3.1 LLM01: Prompt Injection & Adversarial Jailbreaking

• Auctoric Responsibility: Implement real-time multi-stage injection classifiers, inspect semantic embedding anomalies, and isolate system prompts from user prompt execution blocks.
• Customer Responsibility: Implement input length validation, escape untrusted variables, and architect workflows using least-privilege agent tools (e.g., read-only database connections).

3.2 LLM02: Sensitive Information Disclosure

• Auctoric Responsibility: Offer automated client-side and edge PII tokenization/masking filters before inference dispatch; guarantee zero model retention for public training.
• Customer Responsibility: Classify sensitive datasets prior to ingestion; avoid passing unencrypted credentials, proprietary trade secrets, or unmasked PHI into general prompts.

3.3 LLM06: Excessive Agency & Autonomous Tool Execution

• Auctoric Responsibility: Provide deterministic guardrails, function-call permission scopes, runaway loop ceilings, and rate limits on tool execution.
• Customer Responsibility: Enforce human approval gates before downstream agents execute irreversible actions (e.g., executing financial wires, deleting database records, issuing formal legal notices).

4. Compliance RACI Matrix

• R (Responsible): The party executing the activity.
• A (Accountable): The party with ultimate decision ownership.
• C (Consulted): The party providing input or technical resources.
• I (Informed): The party kept updated on status.

5. Contact & Support

For assistance regarding shared controls, architecture reviews, or compliance scoping:

• Security & Compliance Engineering: security@auctoric.ai
• Trust & Security Portal: https://trust.auctoric.ai