AI Service-Specific Terms (SST) — Schedule 1
Document Reference: AUC-LEG-SST-001
Version: 1.2.0
Effective Date: October 24, 2026
Last Updated: October 24, 2026
Classification: Public / Legal Schedule
Incorporated By: Master Services Agreement | Terms of Service
Harmonized Provider Standards: Google Cloud Vertex AI Service-Specific Terms | Microsoft Azure OpenAI Terms | Anthropic Commercial Terms | OpenAI Business Terms | 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 executive navigation pursuant to ISO 24495-1:2023. It does not constitute binding legal terms, covenants, or warranties. In the event of any ambiguity, conflict, or inconsistency between this summary and the operative articles below, the operative articles strictly govern and control.
- Purpose of Schedule 1: Operates as a modular schedule to the MSA and ToS, governing AI-specific capabilities including multi-model routing, agent execution, vector RAG, and fine-tuning.
- Anti-Model Distillation: You agree never to use generated outputs to distill, bootstrap, or train competing foundation or generative models.
- Clinical Healthcare Ban: The platform cannot be used for direct automated patient diagnosis, triage, or writing prescriptions; non-clinical administrative workflows (e.g., billing, scheduling) are permitted.
- Autonomous Agent Safeguards: You control tool authorizations and credentials for agents; Auctoric enforces deterministic recursion caps, token ceilings, and execution timeouts.
- Hyperscaler Flow-Downs: Direct incorporation of Google Cloud Vertex AI, Microsoft Azure OpenAI, and Anthropic commercial restrictions across routed endpoints.
High-Intent Procurement FAQ (AEO Direct Answers)
Can enterprise customers use Auctoric outputs to distill or train their own models?
No. Under Section 3.1, outputs may not be used to train, bootstrap, calibrate, or distill any machine learning or foundation model that competes with Auctoric or its foundational infrastructure providers (Google, Azure, Anthropic).
Does an executed HIPAA BAA permit using Auctoric for clinical medical diagnosis?
No. Under Section 4.1, Auctoric strictly prohibits automated clinical diagnosis, patient triage, and prescription management. Neither a customer’s healthcare credentials nor an executed HIPAA BAA authorizes clinical uses barred by upstream foundation providers.
How does Auctoric prevent runaway execution and billing spikes in autonomous agents?
Under Section 5.2, Auctoric enforces deterministic recursion caps, thread-level token ceilings, and automated execution timeouts to prevent infinite loops, buffer exhaustion, or runaway API consumption.
1. Structure, Applicability, and Order of Precedence
These AI Service-Specific Terms (“SST” or “Schedule 1”) govern Customer’s access to and use of all artificial intelligence inference engines, autonomous agent execution runtimes, retrieval-augmented generation (RAG) vector pipelines, and fine-tuning endpoints provided via the Auctoric Governed AI Platform (“Services”).
This Schedule supplements the Master Services Agreement (“MSA”) or Terms of Service (“ToS”). In the event of any conflict between the universal core terms of the MSA/ToS and this Schedule, this Schedule shall govern with respect to AI-specific functionality, model routing, and outputs.
2. Multi-Cloud & Foundation Model Architecture
Auctoric provides an enterprise multi-model routing gateway capable of orchestrating workloads across leading foundation model providers as well as Auctoric’s private fine-tuned checkpoints:
flowchart TD
App["Customer Application / Agent Client"] --> GW["Auctoric Multi-Model Gateway & Guardrail Engine"]
GW --> Router{"Intelligent Model Router"}
subgraph UpstreamProviders["Upstream Infrastructure Providers (Zero Data Retention Routes)"]
Router -->|Route A: Gemini / PaLM| GCP["Google Cloud Vertex AI<br><i>(Google Flow-Down Terms)</i>"]
Router -->|Route B: GPT-4o / O1| Azure["Microsoft Azure OpenAI<br><i>(Microsoft SST & Code of Conduct)</i>"]
Router -->|Route C: Claude 3.5| Anthropic["Anthropic Commercial API<br><i>(Anthropic Commercial Terms)</i>"]
Router -->|Route D: Llama 3 / Mistral| SelfHost["Auctoric Private Cloud Enclave<br><i>(Isolated Dedicated Hardware)</i>"]
end
GCP --> GW
Azure --> GW
Anthropic --> GW
SelfHost --> GW3. Mandatory Universal Flow-Down Restrictions
To ensure operational stability across our hyperscaler infrastructure partnerships, Customer agrees to the following flow-down covenants derived from the most restrictive terms in our stack (cataloged in the Sub-processor Schedule, including Google Cloud Vertex AI, Microsoft Azure OpenAI, and Anthropic Commercial Terms):
3.1 Strict Prohibition on Competing Model Training & Distillation
Customer warrants, covenants, and agrees that Outputs will NOT be used to train, retrain, bootstrap, calibrate, fine-tune, or distantly supervise any artificial intelligence, machine learning, or foundation models that compete directly or indirectly with Auctoric or its foundational infrastructure providers (as cataloged in the Sub-processor Schedule).
- Customer may use Outputs to validate, evaluate, and test existing internal software systems.
- Synthesizing datasets from Outputs specifically to replicate, distill, or extract the model weights, latent capabilities, or reasoning architectures of underlying platform models is strictly prohibited and constitutes a material, non-curable breach of contract.
3.2 Competitive Benchmarking Restrictions
Customer shall not publicly disclose, publish, or disseminate the results of any competitive benchmarking, stress tests, comparative latency evaluations, or performance assessments of the Services or underlying foundation models without prior written mutual approval from Auctoric and the respective upstream model provider.
3.3 Pass-Through Upstream Abuse Suspension
Customer acknowledges that upstream hyperscalers operate automated network telemetry and threat intelligence systems. If an upstream foundational model provider or hyperscaler (as cataloged in the Sub-processor Schedule) issues an abuse notification, safety strike, or rate throttle directly resulting from Customer’s traffic, prompts, or API calls, Auctoric reserves the contractual right to immediately suspend or restrict Customer’s access to the affected model endpoint without prior notice, until the underlying security or policy violation is remedied. Auctoric will provide notice and restoration criteria as soon as practicable.
4. High-Consequence Sector Restrictions
4.1 Healthcare and Clinical Decision-Making Absolute Prohibition
CUSTOMER EXPRESSLY ACKNOWLEDGES THAT THE SERVICES ARE NOT DESIGNED, TESTED, OR CERTIFIED AS MEDICAL DEVICES UNDER THE U.S. FDA, EU MEDICAL DEVICE REGULATION (MDR 2017/745), OR EQUIVALENT INTERNATIONAL REGULATORY REGIMES.
- Clinical Purpose Ban: Customer shall not utilize the Services, directly or indirectly, for automated medical diagnosis, clinical triage, patient-specific treatment planning, emergency medical recommendations, or prescription management. Neither the presence of licensed professional supervision, nor a customer’s regulated healthcare status, nor the execution of a HIPAA Business Associate Agreement authorizes a clinical use prohibited by an underlying foundation model provider (including Google Cloud Vertex AI).
- Administrative Uses Only: Non-clinical healthcare administrative operations (e.g., medical scheduling, administrative intake drafting, billing coding assistance) are permitted where verified by administrative staff prior to finalization.
4.2 Critical Infrastructure & Kinetic Systems
The Services are not fault-tolerant and are not engineered for use in hazardous, fail-safe environments requiring fail-safe performance. Customer shall not integrate Auctoric into nuclear facility operations, air traffic control systems, real-time public water/electrical utility grids, life support systems, or autonomous weapons platforms.
4.3 High-Stakes Legal, Judicial, and Financial Profiling
Under Article 14 of the EU AI Act and GDPR Article 22, any use of the platform to generate evaluations affecting individual legal status, creditworthiness, loan qualification, employment screening, or housing access must maintain mandatory Human-in-the-Loop (HITL) controls with verifiable human override authority.
4.4 Downstream Application Audience & Minors
Customer warrants and covenants that it will not deploy Customer Applications powered by the Services to target, market to, or be intentionally accessed by individuals under eighteen (18) years of age, complying strictly with upstream foundation model provider restrictions.
5. Autonomous Agents and Tool Execution Governance
When Customer configures autonomous agents capable of dynamic multi-turn execution, web browsing, code execution, or database function calling:
5.1 Tool Authorization and Scoping
- Customer is exclusively responsible for provisioning and scoping API credentials, OAuth tokens, and database permissions supplied to autonomous agents.
- Customer warrants that agents will operate under the principle of least privilege (e.g., read-only tokens for querying data, dual-authorization gates for write/delete actions).
5.2 Recursive Loops and Rate Ceilings
Auctoric enforces deterministic runaway execution limits (maximum recursion depth, maximum token expenditure per thread, execution timeouts) to protect customer billing and platform stability. Customer shall not configure autonomous agents to bypass execution loop throttles.
5.3 Responsibility for Agentic Tool Actions
Customer acknowledges that autonomous agents make probabilistic decisions regarding tool selection. Customer is responsible for its configured permissions, supplied credentials, authorized instructions, and approval decisions. This allocation does not relieve Auctoric of its obligations concerning platform-operated isolation, execution limits, or security controls.
6. Retrieval-Augmented Generation (RAG) & Vector Boundaries
- Cryptographic Tenant Segregation: Customer vector indexes, semantic embeddings, and chunked context files are logically partitioned and encrypted using AES-256 with tenant-scoped encryption keys.
- Access Token Verification: The platform verifies authorization headers on every vector similarity query; no cross-tenant query execution is technically permitted.
- Data Ingestion Representations: Customer represents and warrants that it possesses all requisite licenses, authorizations, and consents to ingest external documents, databases, and proprietary corpuses into the Auctoric Knowledge Engine.
7. Model Non-Deterministic Output Disclaimer
CUSTOMER ACKNOWLEDGES THAT GENERATIVE AI SYSTEMS EMPLOY PROBABILISTIC ALGORITHMS AND LARGE LANGUAGE REASONING CHAINS THAT CAN PRODUCE VARIANT, INCOMPLETE, OR FACTUALLY INCORRECT OUTPUTS (“HALLUCINATIONS”) EVEN WHEN PRESENTED WITH IDENTICAL INPUTS. EXCEPT FOR THE EXPRESS COMMITMENTS IN THE CUSTOMER COPYRIGHT & IP PROTECTION ADDENDUM, AUCTORIC DISCLAIMS ANY WARRANTY THAT OUTPUTS WILL BE DETERMINISTIC, FACTUAL, OR UNBIASED. CUSTOMER ASSUMES RESPONSIBILITY FOR INDEPENDENTLY VALIDATING ALL OUTPUTS BEFORE PRODUCTION CONSUMPTION.
8. Provider-Specific Flow-Down Appendices
When Customer’s requests are routed to specific underlying models (cataloged comprehensively in the Sub-processor Schedule), the following terms apply to that specific transaction:
- Google Cloud Vertex AI (Gemini / PaLM): Workloads are subject to the Google Cloud Generative AI Prohibited Use Policy and Vertex AI Service Terms.
- Microsoft Azure OpenAI Service (GPT models): Workloads are subject to the Microsoft Code of Conduct for Azure OpenAI Service.
- Anthropic Commercial API (Claude models): Workloads are subject to the Anthropic Commercial Terms of Service and Anthropic Usage Policy.