AI Agent Risk Acknowledgment

CRHQ enables you to deploy autonomous AI agents that can perform real actions on your behalf — sending emails, modifying documents, accessing databases, calling APIs, managing files, and interacting with third-party services. This is what makes them powerful. It also means they carry real responsibility.

Before deploying your satellite, please read and understand the following.

1. Agents Perform Real Actions

AI agents on CRHQ are not chatbots. They are autonomous workers that execute tasks in real environments using real credentials. When you configure an agent with access to a service, that agent can perform any action that those credentials allow.

Examples of what agents can do when given appropriate access:

• Read and send emails on your behalf via Gmail or other email providers.
• Create, edit, and delete documents in Google Drive, Notion, or other platforms.
• Query and modify data in connected databases.
• Make API calls to external services using credentials you provide.
• Manage calendar events, spreadsheets, and other productivity tools.
• Execute code, install packages, and modify server files within the satellite environment.
• Navigate websites and interact with web applications via browser automation.

2. The Capability–Risk Balance

There is a direct relationship between how much access you give your agents and how much they can accomplish — but also how much risk they carry. This is no different from giving an employee access to company systems: the more access they have, the more productive they can be, but the more damage a mistake can cause.

Every organization needs to find its own balance between capability and risk. Some teams will want agents that can manage email, deploy code, and interact with clients autonomously. Others may prefer to start with read-only access and expand permissions over time as trust is established.

We strongly recommend starting with limited access and expanding incrementally as you become familiar with how your agents behave in your specific environment.

3. You Control What Agents Can Access

CRHQ does not automatically grant agents access to any external service. You explicitly choose what credentials, API keys, and OAuth connections to provide. You can:

• Grant or revoke third-party service connections at any time from your dashboard.
• Choose which agents have access to which skills and credentials.
• Configure agent instructions to limit scope of actions.
• Use OAuth scopes to restrict what agents can do within a connected service (e.g., read-only Gmail access vs. full send permissions).
• Remove or rotate API keys at any time.

4. AI Models Can Make Mistakes

AI models — regardless of provider (Anthropic, OpenAI, Google, or others) — are powerful but imperfect. They can:

• Misinterpret instructions — an agent may understand a task differently than you intended.
• Produce incorrect output — generated content, code, or data analysis may contain errors.
• Take unintended actions — when given broad permissions, an agent might perform actions you didn't explicitly request as part of completing a task.
• Hallucinate information — AI models may generate plausible-sounding but factually incorrect information.

For high-stakes workflows (financial operations, client communications, data deletion), we recommend implementing review steps, limiting permissions, and monitoring agent activity regularly.

5. Your Responsibilities

By deploying a satellite and using CRHQ agents, you acknowledge and accept that:

• You are responsible for what credentials you provide to your agents. If you give an agent access to your production database, you accept the risk that comes with that access.
• You are responsible for configuring appropriate access levels. CRHQ provides the tools to control agent permissions — using them effectively is your responsibility.
• You are responsible for supervising agent behavior. While agents can work autonomously, reviewing their outputs and actions — especially early on — is essential.
• You are responsible for compliance. If your agents interact with third-party services, you must ensure that the actions they perform comply with those services' terms of use and applicable laws.
• You are responsible for your data. Credentials, API keys, and sensitive information you store on your satellite are your responsibility to manage and protect.

6. What CRHQ Provides

Our responsibility is to provide:

• A secure, dedicated, and hardened server environment for your agents to operate in.
• Encrypted storage for credentials and sensitive data.
• The application layer that enables agent creation, configuration, and collaboration.
• Ongoing infrastructure maintenance, security patches, and platform updates.
• Multi-layer backups for disaster recovery.
• Health monitoring and alerting for your server.

7. What CRHQ Does Not Do

• We do not review, approve, or validate the tasks you assign to your agents.
• We do not monitor the content of your agent conversations or outputs.
• We do not verify that your agents' actions comply with third-party service terms.
• We are not liable for actions taken by AI agents on your behalf, including content generated, emails sent, data modified, or service interactions performed.
• We do not provide the AI models themselves — these are third-party services (Anthropic, OpenAI, Google, etc.) with their own terms, capabilities, and limitations.

8. Best Practices

We recommend the following approach to managing agent risk:

• Start small. Begin with limited permissions and expand as you gain confidence in agent behavior.
• Use read-only access first. Where possible, start with read-only credentials before granting write access.
• Review agent outputs. Especially for new workflows, review what agents produce before it reaches clients or production systems.
• Use scoped credentials. Create service accounts or API keys with only the permissions agents need — not your admin credentials.
• Separate environments. Test agent workflows against staging or sandbox environments before connecting production services.
• Monitor regularly. Check agent conversation logs and activity periodically to ensure agents behave as expected.
• Rotate credentials. Regularly rotate API keys and review which services are connected to your satellite.

9. Acknowledgment

By provisioning a satellite and deploying AI agents on CRHQ, you acknowledge that you have read and understood this document. You accept that:

• AI agents can perform real, consequential actions using the credentials and access you provide.
• The level of risk is directly proportional to the level of access you grant.
• You are responsible for determining the appropriate balance of capability and risk for your organization.
• CRHQ provides the infrastructure and tooling; the decisions about what agents can access and do are yours.

This acknowledgment supplements the Terms of Service and Privacy Policy. In case of conflict, the Terms of Service take precedence.