Security, Data & AI Usage

Each customer deployment is separated from other customer environments.

Where agents are hosted by North Motion, customer workloads run in logically isolated environments with separate:

• application configuration
• credentials
• secrets
• storage
• access controls

Customer environments must not share unrestricted access to data, credentials, or internal systems.

Where required, agents can also be deployed directly into infrastructure controlled by the customer.

Customer data remains the property of the customer.

North Motion does not claim ownership of:

• business data
• documents
• source code
• customer records
• advertising data
• conversations
• credentials
• internal company information

Customer data is processed only as required to provide, maintain, secure, troubleshoot, and operate the agreed service.

Agents should access only the information required to perform their assigned tasks.

We aim to avoid unnecessary:

• data collection
• duplication
• storage
• access permissions

An agent should never receive broader access simply because that access is available.

Access follows the principle of least privilege.

Agents receive only the permissions required for their role.

For example, a marketing agent should not have access to source-code repositories unless that access is explicitly required.

Where supported, we use:

• separate service accounts
• scoped API credentials
• workspace permissions
• role-based access
• restricted integrations

Administrative access to agent infrastructure is limited to authorised personnel.

API keys, access tokens, passwords, OAuth credentials, private keys, and other secrets should not be stored directly inside prompts, source code, public repositories, or general conversation history.

Secrets should be stored using appropriate secret-management or environment-management systems.

Credentials should always use the minimum permissions required for the agent to operate.

Credentials can be rotated or revoked when necessary.

Agents may use external AI providers such as:

• OpenAI
• Anthropic
• Google
• other approved providers

Only information required to complete a given request should be sent to the selected model provider.

The specific provider used may depend on:

• the deployment
• customer requirements
• model capabilities
• privacy requirements
• data residency requirements

Customers can request restrictions on which AI providers may be used.

North Motion does not use customer confidential information to train a proprietary general-purpose AI model unless explicitly agreed otherwise.

Third-party AI providers may apply their own data retention and processing policies depending on:

• provider
• account type
• API plan
• enterprise agreement
• configuration

For sensitive business deployments, enterprise or API-based arrangements with appropriate data controls should be preferred.

Where agents process personal data, processing must have an appropriate purpose and legal basis.

Agent deployments should follow principles such as:

• data minimisation
• purpose limitation
• access control
• retention limitation
• transparency
• security

Depending on the relationship, the customer may act as the data controller and North Motion may act as a data processor.

Where required, this relationship should be documented through a Data Processing Agreement.

Agent deployments may rely on third-party providers for:

• hosting
• databases
• AI models
• communication tools
• observability
• automation platforms
• storage

These providers may act as subprocessors.

Where required, deployments can be designed around specific hosting regions or approved providers.

Customers with strict data residency, contractual, or regulatory requirements should disclose them before deployment.

Where supported by the selected infrastructure:

Data in transit should be protected using encrypted communication such as TLS.

Stored data should use encryption provided by the selected infrastructure where available.

Sensitive credentials should remain separated from ordinary application data.

Agent actions may be logged for:

• troubleshooting
• monitoring
• security
• accountability
• workflow review

Logs may include:

• execution timestamps
• tools used
• workflow status
• errors
• approval events
• agent actions

Sensitive information should be excluded or redacted from logs where possible.

Data should not be retained longer than necessary.

Retention periods may differ between:

• operational databases
• agent memory
• logs
• backups
• AI providers
• connected systems

Customer-specific retention rules can be defined where required.

Persistent agent memory should be treated separately from temporary conversational context.

Agents should retain information only when it is useful for the agreed workflow.

Sensitive information should not be stored in long-term memory unless necessary and authorised.

Customers may define what information an agent is allowed to retain.

AI agents are not inherently infallible.

They may:

• misunderstand instructions
• generate incorrect information
• select the wrong action
• receive incorrect data from connected systems

For consequential actions, deployments should include safeguards such as:

• approval steps
• spending limits
• role restrictions
• allowlists
• environment restrictions
• escalation rules
• execution limits

Higher-risk actions should receive stronger controls.

Agents can operate with different levels of autonomy.

Advisory

The agent prepares recommendations but does not execute actions.

Approval required

The agent prepares work and waits for human confirmation.

Controlled execution

The agent may execute actions within predefined permissions and limits.

Autonomous operation

The agent operates independently inside explicitly defined boundaries.

The level of autonomy should match the risk of the workflow.

Agents may be able to:

• send emails
• send Slack or Teams messages
• publish content
• create advertisements
• update customer records
• create pull requests
• communicate with external users

External communication permissions should be explicitly defined during onboarding.

Where appropriate, agents can be restricted to drafts or internal communication until approval is provided.

Agents with access to advertising, commerce, billing, or financial systems should operate within predefined limits.

These controls may include:

• campaign budgets
• daily spend limits
• account restrictions
• approval thresholds
• restricted action types

North Motion does not recommend unrestricted financial autonomy without additional safeguards.

Agents with access to repositories or infrastructure should use scoped credentials and isolated execution environments.

Where possible:

• changes should use branches
• pull requests should be preferred
• production access should be restricted
• testing should remain in place
• CI/CD controls should remain in place

AI-generated code should not automatically be assumed to be secure or correct.

Where technically appropriate, agents should execute code, process files, and perform automation inside dedicated environments instead of employee workstations.

This helps reduce unnecessary access to local devices and internal systems.

Isolation reduces risk, but does not remove the need for proper permissions and controls.

Agents may process untrusted inputs such as:

• emails
• websites
• uploaded files
• documents
• support tickets
• user-generated content

External content may contain malicious instructions intended to manipulate an AI system.

External content should be treated as data, not automatically as trusted instructions.

Sensitive workflows may require:

• restricted tools
• instruction hierarchy
• allowlisted actions
• validation
• human approval
• execution boundaries

Customers are responsible for determining which systems and information an agent may access.

Customers should:

• provide lawful access
• avoid unnecessary permissions
• notify North Motion of compromised credentials
• define actions requiring approval
• disclose relevant regulatory requirements
• ensure they have the right to process data supplied to the agent

If North Motion becomes aware of a security incident affecting customer systems or data under our control, we will investigate the incident and take reasonable steps to contain it.

Where appropriate, actions may include:

• revoking credentials
• rotating keys
• suspending integrations
• limiting agent access
• investigating affected systems

Affected customers should be notified where required by applicable legal or contractual obligations.

Agent infrastructure, dependencies, integrations, and access controls should be maintained based on risk.

Security actions may include:

• software updates
• dependency updates
• credential rotation
• configuration changes
• integration suspension
• vulnerability remediation

No internet-connected system can be guaranteed to be completely immune from vulnerabilities or attacks.

Where deployments include persistent customer data, backup arrangements may be configured based on the agreed infrastructure and requirements.

Backup configuration may differ by deployment.

Unless explicitly covered by a service-level agreement, uninterrupted availability should not be considered guaranteed.

AI-generated outputs may contain errors, incomplete information, or incorrect assumptions.

Agents should not be used as the sole source of professional:

• legal advice
• medical advice
• financial advice
• compliance decisions
• other regulated decisions

unless the deployment has specifically been designed and approved for that purpose.

Some deployments may fall within obligations under the EU AI Act.

Requirements depend on:

• the type of AI system
• intended use
• the organisation's role
• potential impact
• applicable risk category

Not every AI agent or automation is automatically considered a high-risk AI system.

Classification should be assessed based on the actual use case.

Additional review should take place before deploying agents for areas such as:

• hiring
• employee evaluation
• credit decisions
• healthcare
• biometric processing
• critical infrastructure
• legal decision-making
• access to essential services
• high-impact decisions about individuals

North Motion may restrict or reject deployments where appropriate safeguards cannot reasonably be implemented.

Agents may process confidential business information where required for their task.

Access should be limited to relevant:

• systems
• personnel
• providers
• workflows

Customers should identify especially sensitive information during onboarding so additional controls can be applied.

Every deployment should be built around six core principles:

Isolation

Customer environments remain separated.

Least privilege

Agents receive only the access they need.

Data minimisation

Only required information should be processed.

Controlled autonomy

Higher-impact actions receive stronger safeguards.

Traceability

Relevant actions can be logged and reviewed.

Human oversight

People remain involved where judgment or risk requires it.