Skip to content

Governance

Your data, your AI agents and the laws that govern them

Who is responsible for what when your company hands work to agents, what is in place in every Belowdecks workspace and what we can adapt to your requirements, like hosting in Quebec.

Who is responsible for what

Everyone’s role

Law 25 and the GDPR distinguish the company that decides how personal information is used from the service provider that processes it on its behalf. Here is how these roles are shared with Belowdecks.

  1. Step 1 of 4: A customer writes to you. Who acts: The people concerned, Your customers and employees
  2. Step 2 of 4: The agent reads the request in its folder. Who acts: Agents
  3. Step 3 of 4: The AI provider processes the text. Who acts: Your AI provider, Anthropic or OpenAI
  4. Step 4 of 4: Your team approves the reply. Who acts: Responsible for the information, Your company
    • Exception. Incident: notice without delay to your privacy officer. Who acts: Service provider, Belowdecks team
  • The people concerned, Your customers and employees

    Outside,Writes to your company

    Their information goes through the agents when they write to you, order or apply for a job. The law gives them the right to access it, have it corrected and withdraw their consent.

    • Exercise their rights with you
  • Responsible for the information, Your company

    In your company,Decides

    You decide which tasks go to the agents, which information they use and which actions wait for a person’s approval. You remain responsible for this information, as with any other software.

    • Decides how it is used
    • Names the person in charge
  • Agents

    In Belowdecks,Works

    Each agent reads only its own work folder and the software you connect to it. For the actions you choose, it prepares the work and waits for the approval of a person on your team.

    • Limited to their folder
    • Wait for your approval
  • Your AI provider, Anthropic or OpenAI

    Outside,Sends an event

    The provider processes the text the agents send it, with an account in your company’s name and under its own terms. These terms state where it processes this text and how long it keeps it.

    • Account in your name
    • Under its own terms
  • Service provider, Belowdecks team

    Behind the scenes,Sets up and monitors

    We install, host and maintain your workspace following your instructions and the agreement signed with you. We use your information only for this mandate, and we notify you of an incident without delay.

    • Hosts and maintains
    • Notifies you of an incident

In place today

What every Belowdecks workspace includes

These measures come from the software or from the way our team sets up each workspace. The Security page describes them in detail.

  • One database per company

    Your workspace has its own database and its own accounts. No information passes from one client company to another.

  • Limited access

    Each agent sees only its own work folder, and each member of your team has only the permissions of their group. Passwords and keys are encrypted and masked in the logs.

  • A person’s approval before important actions

    We choose with you the actions that wait for the approval of a person on your team, like writing to a customer or changing your books.

  • A record of every task

    Every task leaves a summary of what the agent did, its files and its result. You can stop a running task or pause everything with a single button.

  • Belowdecks trains no model

    We do not use your data to train an AI model. What your AI provider may do with the data it receives depends on its terms, linked in the questions below.

  • The list of what leaves your workspace

    The Security page names every piece of data that leaves your workspace and where it goes: the AI provider, the software you connect, email delivery, error tracking and the backup copy.

  • A backup every night

    The database and the files are archived every night, and a copy goes to separate storage.

  • A workspace maintained by our team

    We apply your workspace’s updates and we track its technical errors.

On request

What we can adapt to your requirements

These options are decided during the diagnostic, based on how sensitive your information is and on your obligations. They are part of the quote.

Hosting in Quebec

We can install your workspace and its backups in a data center in Quebec, with a hosting provider chosen with you at the diagnostic.

The AI provider of your choice

The agents can work with Anthropic, OpenAI or other providers, such as Mistral AI. We read their terms with you: where each one processes data and how long it keeps it.

An open AI model on a server in Canada

An open AI model can run on a server in Canada, with no outside AI provider. This option needs GPUs, and we study it project by project.

A written processing agreement

We sign an agreement with you that sets out our protection measures, use limited to our mandate, incident notice, your right to verify and the destruction of the information at the end of the contract.

Retention set to your policy

We set how long the task history and the backups are kept to match your retention and destruction policy.

The documents for your assessment

We give you a description of your workspace’s data flows, its providers and its security measures for your privacy impact assessment.

The laws

The laws that govern your agents

Depending on where your customers and employees are, several laws can apply at the same time.

Questions we get about governance

Who is responsible if an agent makes a mistake with personal information?

Your company remains responsible for the personal information it holds, including the information the agents process. That is why we set with you the actions that wait for a person’s approval, and every task leaves a record of what the agent did. The agreement signed with us sets out our own obligations as a service provider.

Can my data stay in Canada?

Your workspace and its backups can be hosted in Quebec. The text the agents send to the AI provider is processed where that provider states in its documentation, for example Anthropic’s or OpenAI’s.

For no text to leave Canada, you need an open model installed on a Canadian server, an option we study project by project.

Can an agent make a decision about a person on its own?

It can if you set it up that way. Law 25 then requires you to inform the person of this decision based exclusively on automated processing and to let them have it reviewed by a member of your team (section 12.1). For decisions that affect a person, like hiring or credit, we recommend that a person on your team decides.

What happens to my data at the end of the contract?

You export your settings and your lists, and your AI provider account stays in your name. We then delete your workspace and its backups within the time set in the agreement.

Does Belowdecks hold a security certification?

No. Belowdecks does not hold a certification like SOC 2 or ISO 27001. AI providers publish their certifications in their trust center, for example Anthropic’s or OpenAI’s. Our measures are described on the Security page, and we answer your security questionnaires during the diagnostic.

Where can I read the AI providers’ terms?

Each provider publishes its own terms: where it processes data, how long it keeps it, whether it may use it for its models and which certifications it holds. The provider can change these documents.

First call

Let’s talk about your obligations before choosing the first task

Twenty free minutes to see whether Belowdecks fits your company. The diagnostic then describes where your data will be hosted and what leaves it.