AI and Automated Actions Policy

Effective and last updated: July 28, 2026

Merchant Command uses artificial intelligence and software agents to analyze business information, recommend actions, and, when authorized, take actions through connected services. This Policy explains how those systems work, how authorization is treated, and the responsibilities that remain with Customers.

1. Purpose and Scope

This Policy explains how Merchant Command uses artificial intelligence and software agents, how approvals and standing permissions work, and what Customers must do to use automated actions responsibly. It applies to all AI-generated Output, agent workflows, tool calls, Connected Service activity, and automated actions through Merchant Command.

This Policy supplements and is incorporated into the Terms of Service. If this Policy conflicts with the Terms, the Terms control unless this Policy expressly provides a more specific rule for AI or automated actions.

2. What Merchant Command Agents Do

Merchant Command may combine generative AI models, classifiers, retrieval systems, deterministic rules, code, APIs, and other tools. An Agent is a software process that can interpret instructions, analyze information, select tools, and perform one or more operations toward a task.

Depending on permissions and configuration, an Agent may:

• retrieve and summarize data from Shopify or another Connected Service;
• compare performance, identify patterns, and recommend actions;
• draft product content, campaigns, messages, reports, or operational changes;
• create, update, publish, send, refund, cancel, or otherwise act through a Connected Service; and
• sequence multiple tool calls or collaborate with other agents.

Agents are not people, employees, fiduciaries, licensed professionals, or independent decision-makers. They operate through software, model outputs, rules, and Customer-granted authority.

3. Multiple Models and Providers

Merchant Command uses different AI models and providers for different tasks. Depending on the route and feature, models or services may be developed, hosted, or supplied by OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral AI, xAI, Alibaba/Qwen, Perplexity AI, and others.

The model developer may not receive Customer data when a model is self-hosted or hosted by another provider. Merchant Command may select or change a model based on task fit, capability, cost, latency, availability, safety, Customer preference, or operational needs. Model names, versions, and providers may change without a separate amendment when the change is not materially less protective of Customer data.

No single model is guaranteed for any feature. Different models can produce materially different answers or actions from the same instruction.

4. Data Sent to AI Systems and Model Training

To perform a task, Merchant Command may send an AI system the user's prompt, system instructions, relevant chat history, selected files, Connected Service data, tool results, and other context reasonably needed for the task. We seek to limit the context to what is relevant, but AI workflows may process personal, confidential, or commercially sensitive information contained in Customer Data.

Our standard is to use business, enterprise, API, self-hosted, or other configurations that do not permit Customer Content to train generalized AI models. Mivaro does not use Customer Content or Connected Service data to train generalized models or improve models for other customers unless Customer separately and affirmatively opts in.

We do not intentionally route Customer Content to a provider under terms that permit generalized model training unless Customer gives separate affirmative authorization. If a provider cannot meet that standard, we may restrict its use to public or non-sensitive information, use a different hosting arrangement, or disable the route.

AI providers may retain prompts, outputs, and metadata for limited periods for service delivery, safety, security, abuse prevention, debugging, or legal compliance. Limited human review may occur under provider terms. Mivaro personnel may access relevant data only as described in the Privacy Policy, including when a user reports an issue, grants temporary support access, or access is needed for security or maintenance.

Do not place unrelated credentials, full payment-card data, Social Security numbers, medical records, or other prohibited sensitive information in prompts or files. Authentication secrets should be supplied only through designated connection controls.

5. Operation and Approval Categories

Merchant Command may classify an operation by its context and risk. Typical categories are:

5.1 Read and analysis operations

These operations retrieve, monitor, search, compare, summarize, or analyze data without intentionally changing a Connected Service. They often do not require per-action approval. A read can still expose sensitive data, incur third-party costs, or create risk, so Merchant Command may require approval or limit access in some circumstances.

5.2 Proposed or draft actions

An Agent may prepare a proposed change, message, campaign, refund, listing, or workflow without executing it. The proposal remains subject to Customer review unless the Customer has separately authorized execution.

5.3 Approval-required actions

These are actions that Merchant Command presents to an Authorized User for confirmation before execution. Approval may include reviewing a preview, selecting a scope, entering a confirmation, or using another control.

5.4 Pre-authorized actions

These are actions approved in advance through an account setting, agent instruction, workflow, rule, scope, monetary threshold, frequency limit, recipient restriction, or similar control. An action within that scope is treated as Customer-authorized even if no person approves it at execution time.

An operation's category may vary by amount, destination, frequency, reversibility, data sensitivity, platform, and Customer settings. Mivaro may reclassify, block, or require additional approval for an action at any time to protect security, safety, legal compliance, providers, or Customers.

6. Standing Permissions and Customer Authority

Account owners and administrators are responsible for deciding who may create agents, connect services, approve actions, and grant standing permissions. Customer is bound by permissions and approvals configured by its Authorized Users within their roles.

Where controls are available, Customers should use least-privilege scopes, narrow data access, monetary limits, frequency limits, approved recipients, review thresholds, and expiration dates. Broad instructions such as "handle everything automatically" can create substantial risk and should be avoided unless Customer has tested the workflow and understands the consequences.

Standing authorization remains effective until revoked or changed. Revocation is prospective. An action already submitted to a Connected Service, placed in a queue, sent to a recipient, or otherwise completed may not be stopped or reversed.

Customer is responsible for charges, usage, transactions, advertising spend, messaging fees, API fees, infrastructure costs, refunds, discounts, purchases, and other financial consequences generated within permissions Customer grants, including consequences of repetitive, recursive, misconfigured, or unexpectedly high-volume workflows. Customer should configure budgets, quotas, rate limits, approval thresholds, alerts, and automatic stopping conditions where available.

Merchant Command may impose non-delegable approvals, spending caps, rate limits, restricted destinations, or other safeguards even when Customer requests broader autonomy. A safety control does not shift responsibility from Customer and does not guarantee that harmful or unintended activity will be prevented.

Mivaro may pause or limit activity that appears anomalous or excessive, but does not guarantee that it will identify or prevent unintended activity or costs before they occur.

7. External and Higher-Impact Actions

Depending on integrations and permissions, Agents may update product titles, descriptions, images, prices, discounts, inventory, orders, fulfillment, advertising campaigns, budgets, customer records, or other business data; publish content; issue refunds; and send email, SMS, support, or other communications.

These actions can affect revenue, costs, customer relationships, legal obligations, platform standing, and third parties. Customer must apply a level of human review appropriate to the potential impact.

Customer is responsible for:

• verifying recipients, amounts, prices, quantities, claims, dates, and material facts;
• ensuring communications have required consent, identification, disclosures, and opt-out mechanisms;
• complying with refund, return, advertising, consumer-protection, privacy, and marketplace rules;
• setting limits for spending, discounts, refunds, and repetitive actions;
• reviewing proposed changes before broad publication or deployment; and
• maintaining backups or exports of critical data before enabling writes or deletion.

Some actions may remain approval-required or unavailable regardless of Customer settings. Mivaro may block a transaction, communication, deletion, account change, or other action that appears unsafe, unlawful, fraudulent, outside scope, or inconsistent with a provider's rules.

8. Required Human Oversight

Customer must maintain meaningful human oversight appropriate to its use case. At a minimum, Customer should:

• define the Agent's task, authority, boundaries, and escalation rules;
• test workflows with limited scope before production use;
• review Output for accuracy, completeness, bias, tone, and legal compliance;
• monitor logs and investigate unexpected behavior;
• periodically review connections, scopes, standing permissions, and Authorized Users;
• provide a human escalation path for customer-facing interactions; and
• stop or narrow automation when errors, drift, abuse, or changed circumstances appear.

During free, trial, beta, pilot, or other evaluation access, Mivaro may require approval-required settings for write operations or other higher-impact actions and may restrict standing permissions. Where approval controls are available, Customer should keep them enabled until the workflow has been tested with limited scope. If Customer enables standing permissions or reduces approval checkpoints, Customer assumes responsibility for actions taken within those settings, subject to the Terms of Service.

Customer may not use Merchant Command as the sole basis for legally or similarly significant decisions about individuals in employment, housing, credit, insurance, healthcare, education, legal services, government benefits, or other essential services. Any permitted use in those contexts requires qualified human review and compliance with the Acceptable Use Policy.

Output is not professional advice. A qualified person must review and adopt any legal, tax, accounting, investment, medical, employment, or regulatory conclusion before Customer relies on it.

9. Known AI and Agent Risks

Customers should expect and plan for the following risks:

• Hallucination and error: a model may invent facts, citations, calculations, product details, or policies.
• Outdated or incomplete context: an Agent may not have current or complete data.
• Bias and inconsistency: models can produce unequal, unstable, or inappropriate results.
• Prompt injection: external content, webpages, messages, product data, or files may contain instructions designed to manipulate an Agent into disclosing data, ignoring safeguards, or performing unauthorized actions.
• Tool-selection error: an Agent may call the wrong tool, use incorrect parameters, or misunderstand an API response.
• Permission error: a connection may grant broader access than intended, or a user's role may be misconfigured.
• Third-party failure: APIs may be unavailable, delayed, rate-limited, changed, or return incomplete data.
• Irreversibility: messages, publications, deletions, refunds, and other actions may not be reversible.
• Duplicate or partial execution: retries, timeouts, or provider errors may cause duplicated, delayed, or partial actions.
• Runaway or recursive execution: a misconfigured, compromised, or manipulated workflow may repeat, amplify, or escalate actions beyond the intended scope.
• Cost amplification: repeated model calls, messages, ads, refunds, purchases, API operations, or other actions may generate unexpected or substantial charges.
• Data exposure: prompts, outputs, logs, files, and tool results may contain confidential or personal information.

Merchant Command may use safeguards, filters, previews, validation, confirmations, and monitoring, but no safeguard can eliminate these risks.

10. Logs, Records, and Support Review

Mivaro maintains records of material agent activity for operational, security, audit, support, and dispute purposes. Records may include prompts, outputs, data sources, tool calls, approvals, standing permissions, changes, actors, timestamps, errors, and provider metadata.

Logs may not capture every internal model step or every event within a Connected Service. Model reasoning may not be available, and a displayed explanation may be a summary rather than a complete account of internal computation.

When a user reports an issue, the report may share relevant conversations, outputs, logs, screenshots, or agent activity with support. A user may also provide a temporary support code or permission for account access. Authorized personnel may access data without a support code when necessary for security, maintenance, incident response, abuse investigation, or legal compliance.

11. Output Rights and Publication

As between Mivaro and Customer, Customer may use and owns Output to the extent permitted by law and subject to third-party rights. Output may not be unique, copyrightable, accurate, or noninfringing. Other customers may receive similar material.

Before publishing or sending Output, Customer must review it for confidential information, personal information, false claims, infringement, defamation, required disclosures, and compliance with the destination platform's rules.

Customer may not represent that Mivaro has endorsed, certified, verified, or guaranteed an Output or action unless Mivaro expressly states so in writing.

12. Prohibited and Restricted Uses

AI and automated actions must comply with the Acceptable Use Policy. Among other restrictions, Customer may not use them for unauthorized access, fraud, spam, unlawful surveillance, harmful or illegal activity, fake reviews, deceptive claims, rights infringement, or prohibited high-impact decision-making.

Customer may not intentionally circumvent safeguards or use prompt injection, adversarial instructions, or tool manipulation to obtain unauthorized data or cause an Agent to exceed approved permissions.

13. Responding to Unexpected or Harmful Activity

If an Agent behaves unexpectedly or performs an incorrect action, Customer should promptly:

• pause or disable the Agent, workflow, and related downstream triggers;
• revoke or narrow standing permissions and Connected Service scopes;
• rotate credentials if compromise is suspected;
• review logs and the Connected Service's native activity history;
• check for repetitive, recursive, anomalous, or unexpectedly high-volume execution and associated charges;
• reverse or correct the action where possible;
• preserve relevant evidence;
• notify affected people or providers where required; and
• contact support@merchantcommand.com with relevant details.

Mivaro may block the specific action, pause or disable an Agent or workflow, narrow or revoke a connection, require reauthorization, restrict the account, or suspend or terminate access while investigating. We do not guarantee that we can reverse an action taken through a third party.

14. Changes and Contact

We may update this Policy as models, providers, features, laws, and risks change. Material changes will be communicated as described in the Terms of Service. We will not retroactively authorize generalized model training on previously collected Customer Content without separate affirmative consent.

Questions or concerns:

Mivaro Labs LLC

support@merchantcommand.com