> For the complete documentation index, see [llms.txt](https://legal.lampi.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://legal.lampi.ai/security-measures-notice.md). # Security Measures Notice | At Lampi AI, security and privacy by design are core principles in the delivery of our AI-related services. We take seriously our responsibility to protect the software, systems, infrastructure, and data that support the Services, and we maintain technical and organizational measures intended to preserve the confidentiality, integrity, and availability of Customer Data and Personal Information. | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | This Security Measures Notice forms part of, and is subject to, the Master Services Agreement, Data Processing Addendum, subscription agreement, or other written or electronic terms between Lampi AI and the Customer that reference this notice (the “**Agreement**”). It is effective as of the effective date of the Agreement, unless otherwise agreed by the Parties. This notice describes the main security measures and controls implemented by Lampi AI to protect Customer Data and Personal Information in connection with the Services. It is intended to provide a clear and trust-building overview and does not disclose confidential implementation details, internal configurations, or security-sensitive operational procedures. Additional information regarding Lampi AI’s security measures may be made available upon request and subject to appropriate confidentiality, security, and contractual safeguards. Lampi AI may update these technical and organizational measures from time to time, including to reflect changes in technology, regulatory expectations, service architecture, or security practices, provided that such updates do not materially reduce or weaken the overall level of protection for Personal Information processed by Lampi AI in providing the Services. Customer acknowledges that it is responsible for reviewing the information made available by Lampi AI, including this notice, and determining whether the Services and related security measures meet Customer’s business requirements, internal policies, and legal obligations under applicable laws. Capitalized terms not defined in this Security Measures Notice have the meanings given to them in the Agreement or the Data Processing Addendum. ### 1. Technical measures **Data hosting and infrastructure.** Except as otherwise agreed by the Parties, Lampi AI hosts its production platform primarily on Microsoft Azure, with core workloads deployed as containerized services on Azure Kubernetes Service (AKS) in the European Union, including France as the default production hosting location where applicable. The platform uses Azure-native storage, networking, compute, and security capabilities to support resilient and controlled operation of the Services. **Architecture overview.** Lampi AI’s platform is designed as a set of separated application services and microservices, including frontend services, backend APIs, AI processing services, inference components, document ingestion and retrieval workflows, integration bridge components, GPU-related workloads, and orchestration, autoscaling, and scheduling layers. This separation supports controlled deployment, scalability, monitoring, and isolation of responsibilities across the platform. **Private-by-default infrastructure.** Lampi AI applies a private-by-default service exposure model for internal platform components. Databases, search engines, cache and state services, internal processing services, logging components, and storage backends are not designed to be directly exposed to the public internet. Production services are exposed only through designated and controlled entry points. **Data storage and processing.** Lampi AI separates data storage and processing components by function, including document and object storage, relational data storage, vector and full-text search, cache or state storage, AI workflow state, logs, and generated artifacts. Full-text and vector-enabled search capabilities are supported through Elasticsearch, while uploaded and generated files are stored through S3-compatible object storage, such as MinIO, hosted within the Azure environment where applicable. **Network segmentation**. Lampi AI uses AKS networking, Kubernetes service boundaries, internal service discovery, and controlled ingress routes to segment production workloads. External HTTP(S) traffic is routed through designated ingress components, such as NGINX HTTPS ingress where applicable, while internal services communicate through private cluster networking and are subject to service-level access controls. **Administrative access points.** Administrative and monitoring interfaces are restricted to authorized personnel and protected through authentication and network controls such as VPN, IP allowlisting, or equivalent access restrictions where applicable. Direct public access to production databases, internal object storage, caches, search services, and internal processing components is not permitted. **Encryption in transit.** Lampi AI protects communications in transit using HTTPS/TLS, with TLS 1.3 used where supported by the relevant component, provider, or integration. External access to files or objects is handled through controlled mechanisms such as time-limited presigned URLs where applicable, rather than direct public access to storage backends. **Encryption at rest.** Customer Data and Personal Information are protected at rest using AES-256 or cloud-provider managed encryption mechanisms, including for databases, object storage, managed disks, managed file storage, and backup storage where applicable. Encryption controls may be implemented through the application, infrastructure, database, or cloud provider layer depending on the data class and service component. **Secrets and key management.** Lampi AI restricts access to secrets, credentials, and cryptographic material using role-based access, least-privilege permissions, infrastructure secret stores, and Kubernetes secrets or equivalent mechanisms where applicable. Administrative access to sensitive systems requires appropriate authorization and multi-factor authentication. Application keys and service credentials are reviewed and rotated periodically according to operational requirements and risk. **Authentication methods.** Lampi AI supports authentication methods such as email and password where enabled, signed-link password recovery, Azure AD / Microsoft Entra ID single sign-on using OAuth2 or OpenID Connect, and Google OAuth where configured. Passwords are not sent in plain text, and password-based authentication uses secure password hashing mechanisms such as bcrypt. **Multi-factor authentication**. Lampi AI requires or enforces multi-factor authentication for sensitive administrative and infrastructure access where applicable. The platform includes protections against brute-force attacks, including account lockout or blocking mechanisms after repeated failed authentication attempts, as configured for the relevant account type or service. **Role-based access control.** Lampi AI implements role-based access control to restrict access to features, data, administrative functions, and operational capabilities according to assigned roles and permissions. Access rights are designed according to the principle of least privilege and are enforced through backend authorization checks, with user interface controls used as an additional layer where applicable. **Customer role-based access.** To limit access based on the principle of least privileged access and help prevent conflicts of interest, Customer administrators can manage users, roles, and permissions within the Services. This may include restricting access to features, workspaces, data sources, administrative capabilities, integrations, and available large language models based on users’ responsibilities, seniority, or business need. **Tenant isolation.** Lampi AI enforces tenant isolation through organization-scoped authorization, path and object segregation, API-level validation, and membership and permission checks on relevant requests. Client access to object storage is mediated through controlled application flows and time-limited links where applicable. **API and service access**. API access and service-to-service communication are protected through authentication, authorization, and least-privilege service credentials where applicable. Internal APIs and backend services are not intended to be publicly reachable except through approved ingress, gateway, or application routes. **Upload and file controls**. Lampi AI applies security controls to uploaded content, including file-type validation and malware scanning or equivalent protective mechanisms where applicable. These controls are intended to reduce the risk of malicious files, unsupported formats, and unsafe processing behavior within document ingestion and AI workflows. **Logging and monitoring**. Lampi AI uses monitoring, tracing, and logging tools. Logs and telemetry may include application logs, infrastructure events, Kubernetes events, access and authentication events, configuration activity, administrative actions, operational metrics, AI workflow traces, and data-operation audit events, subject to configured retention and the Agreement. **Log protection**. Logs are stored in internal logging and monitoring systems and are not directly exposed to the public internet. Access to logs is restricted to authorized personnel through role-based permissions, authentication, and infrastructure access controls. Log retention periods may vary by log type, service component, operational need, and contractual commitment. **Vulnerability and patch management**. Lampi AI performs vulnerability scanning, dependency monitoring, container image scanning, and package monitoring across its infrastructure-related hosts, codebase, and product components. Identified issues are triaged and remediated according to severity, exploitability, affected component, and operational risk. **Patch deployment and verification**. Lampi AI manages patches through controlled release and deployment processes, including version-controlled changes, container rebuilds where applicable, CI/CD testing, and post-deployment monitoring. Emergency or critical security patches may be expedited, subject to appropriate validation and rollback procedures. **Secure development.** Lampi AI follows a secure development lifecycle. Secure coding practices are aligned with widely recognized principles such as the OWASP Top 10 and are adapted to the specific risks of AI-enabled applications. **Testing and non-production environments**. Lampi AI maintains development, staging, and test environments that are separated from production through distinct configurations, deployment stacks, access controls, domains or endpoints, and CI/CD execution paths where applicable. Real Customer Data is not intended to be used in test environments; public, synthetic, anonymized, or otherwise approved data is used for testing where appropriate. **Backup and resilience**. Lampi AI performs scheduled backups of relevant production components, such as databases, search systems, object storage, cache or state components, and tracing or observability systems, according to the applicable service configuration. Backups are stored in Azure-hosted storage within the European Union perimeter where applicable and are protected in transit and at rest using cloud-provider encryption and secure transfer mechanisms. **Restore testing and disaster recovery**. Lampi AI maintains backup restoration and disaster recovery procedures covering data stores, application configuration, and operational recovery steps. Restore testing is performed at least annually or as otherwise required by the Agreement, and disaster recovery procedures include escalation, technical recovery actions, role assignment, and customer communication where applicable. **Physical security**. Lampi AI relies on reputable cloud infrastructure providers for physical security controls protecting data centers, hardware, facilities, power, environmental protections, and physical access management. These controls are managed under the shared responsibility model and supported by the providers’ security and compliance programs. ### 2. Organizational measures **Security governance**. Lampi AI maintains an internal governance structure for information security and operational resilience. The CEO provides executive oversight and accountability for security and resilience priorities, while the CTO has operational responsibility for implementing and maintaining security controls, incident response procedures, and technical risk management practices. **Security team responsibilities.** Lampi AI’s Security Team monitors systems, reviews security events, coordinates incident response, supports vulnerability management, and assists with containment, investigation, remediation, and documentation of security events. Security responsibilities are assigned according to role, expertise, and operational need. **Employee awareness and training**. Lampi AI provides onboarding and periodic security and privacy awareness training to personnel who work with systems, data, or operational processes. Training includes expected behavior, reporting channels, confidentiality obligations, data protection principles, secure use of internal tools, and awareness of common threats such as phishing and social engineering. **Personnel security**. Lampi AI personnel are required to comply with confidentiality, acceptable use, professional conduct, and information security expectations. Personnel are required to execute confidentiality commitments, acknowledge applicable policies, and process Customer Data only as authorized. Background checks may be conducted where legally permissible, role-appropriate, and consistent with applicable labor laws. **Internal access control**. Lampi AI applies least-privilege and role-based access principles to internal systems, cloud infrastructure, production environments, databases, networks, logs, and administrative tools. Access to sensitive systems is restricted to authorized personnel with a legitimate business need and is protected through authentication, multi-factor authentication where applicable, and periodic review. **Privileged access management**. Privileged access is limited to authorized personnel and used only for administration, support, troubleshooting, security, recovery, or other approved operational purposes. Privileged actions are subject to approval, logging, access restriction, and time limitation where appropriate. Access may be revoked, suspended, or restricted immediately where required for security, termination, or contractual reasons. **Access revocation.** Lampi AI maintains onboarding, role-change, and termination procedures designed to ensure that employee and contractor access is granted appropriately and revoked when no longer required. Termination or offboarding processes include revocation of relevant credentials, tokens, system accounts, and administrative privileges. **Risk management**. Lampi AI conducts periodic reviews of security, privacy, operational, and supplier risks. These reviews support continuous improvement of policies, controls, monitoring, incident response procedures, and development practices. Lampi AI’s practices are designed to be aware of applicable requirements and expectations under GDPR, NIS2, DORA, and other relevant frameworks where applicable. **Data protection governance**. Lampi AI processes Customer Data and Personal Information in accordance with the Agreement, the Data Processing Addendum, and Customer instructions where applicable. Lampi AI applies data minimization, purpose limitation, access restriction, encryption, and data residency controls appropriate to the Services and applicable processing activities. **Policy management.** Lampi AI maintains policies and procedures covering security, privacy, incident response, access management, vulnerability management, supplier risk, and business continuity where applicable. Policies and procedures are reviewed periodically and updated to reflect changes in the Services, regulatory environment, threat landscape, and business operations. ### 3. Supplier and third-party controls **Supplier management.** Lampi AI limits supplier relationships relevant to service delivery to providers with an appropriate business purpose and security posture. Suppliers are assessed based on their role, criticality, data access profile, contractual commitments, operational resilience, and available assurance documentation. **Sub-processors.** Lampi AI’s sub-processors may include cloud infrastructure providers, approved AI or large language model providers, and business service providers where applicable. The current list of sub-processors is made available separately in Lampi AI’s Sub-Processors Notice or otherwise in accordance with the Agreement. **Contractual protections.** Lampi AI uses data processing agreements, confidentiality obligations, security clauses, incident notification obligations, data transfer safeguards, audit or information rights, and sub-processing controls where appropriate. Supplier commitments are evaluated in light of the supplier’s role, the nature of the processing, and the level of access to Customer Data. **Cloud provider assurance**. Lampi AI relies on critical cloud and infrastructure providers, including Microsoft Azure as its primary production cloud provider, for portions of the security control environment under the shared responsibility model. Such providers are selected in part for their mature security programs, resilience capabilities, regional hosting options, and recognized compliance documentation. **Third-party AI providers**. Where a Customer or Authorized User elects to use a third-party model provider or AI service through the Services, such use is governed by the configuration, provider terms, contractual safeguards, and Customer instructions applicable to that integration. Lampi AI use only providers and configurations that restrict retention, reuse, and training on Customer Data. **Third-party integrations**. Customer-driven connectors and integrations are permission-based and are designed to access only the resources authorized by the Customer or end user. Integrations do not receive ongoing access beyond the permissions granted, and access can be revoked according to the relevant application and service configuration. **Data residency and transfers**. Customer Data is stored and processed primarily in the European Union by default, subject to Customer configuration, selected integrations, selected model providers, and the Agreement. Cross-border transfers, where applicable, are managed through contractual, technical, and organizational safeguards such as appropriate data processing terms and transfer mechanisms. ### 4. AI-specific security and data protection **AI security by design.** Lampi AI designs its AI-related services with controls intended to protect Customer Data during AI processing, retrieval, model interaction, inference, document analysis, and generated output workflows. These controls are implemented across application logic, authorization checks, infrastructure, logging, model routing, and operational monitoring. **Customer control over model use.** Customers may configure or agree which model providers, model families, or deployment options are available for use within the Services, subject to the Agreement and technical availability. External large language model calls can be restricted or disabled in favor of approved, self-hosted, EU-hosted, private, or on-premise models where supported by the deployment model. **Use of Customer Data.** Lampi AI does not sell Customer Data, prompts, inputs, outputs, or Personal Information, and does not use such data to train or improve Lampi AI models except where the Customer has explicitly requested and agreed to a dedicated fine-tuning or training activity. Where fine-tuning is requested, additional controls, customer approvals, environment separation, and anonymization or deletion of Personal Information are applied where applicable. **Third-party model processing.** When a Customer or Authorized User chooses to use a third-party model through the Services, Customer Data may be transmitted to that provider solely for the purpose of providing the requested functionality and subject to applicable contractual and technical safeguards. Lampi AI uses only customer-selected third-party AI and model providers under commitments that Customer Data, prompts, inputs, and outputs are not used to train or improve third-party models. **Zero data retention and non-use commitments.** Lampi AI uses third-party model provider that apply zero data retention, non-retention, non-logging, or equivalent non-use commitments for Customer Data submitted for inference. These commitments are intended to ensure that Customer Data processed by third-party model providers is not retained, logged, reused, or incorporated into provider model-training datasets. **Model routing and restriction controls.** Lampi AI applies model routing controls to manage which AI endpoints may be called in production and to reduce unintended data exposure to non-approved providers. Customers may choose to restrict available models, disable external model calls, or use approved self-hosted, EU-hosted, private, or on-premise models where applicable to their deployment and contractual requirements. **Retrieval and data-source controls.** Lampi AI’s AI workflows use authorization-aware retrieval and data-source access controls so that users receive responses based on data they are permitted to access. Organization membership, role permissions, connector permissions, and document access rules are validated as part of relevant retrieval and processing workflows. **Prompt-injection safeguards.** Lampi AI applies AI-specific security controls intended to reduce the risks of prompt injection, unsafe instruction following, unauthorized tool use, and unintended data exposure. **AI observability and auditability.** Lampi AI maintains observability and tracing for AI workflows. This supports operational monitoring, debugging, quality review, and auditability of AI interactions, including the relationship between user requests, data sources, model calls, and generated responses, subject to configured retention and access restrictions. **Human oversight.** Lampi AI’s AI capabilities are designed primarily to augment human analysis, document processing, and decision support. Customers remain responsible for their use of outputs, validation of critical decisions, internal governance, and configuration of user permissions and model access according to their own policies and legal obligations. ### 5. Incident response, continuity and availability **Incident response program.** Lampi AI maintains policies and procedures for detecting, escalating, containing, investigating, remediating, documenting, and communicating security and operational incidents. Incidents are classified by category and severity, including a priority scale from P1 to P5, to support appropriate response, escalation, and customer communication. **Detection and escalation**. Lampi AI uses operational monitoring, logs, alerting, employee reporting, vulnerability management, and security tools to identify potential incidents. Personnel are required to report suspected incidents promptly, and high-risk events are escalated to the CTO and Security Team for coordination and decision-making. **Containment and remediation**. Lampi AI’s incident response procedures include containment, investigation, evidence preservation, root cause analysis, corrective action, recovery, monitoring, and post-incident review. Remediation may include configuration changes, patches, access revocation, provider coordination, customer guidance, or procedural improvements depending on the nature of the incident. **Customer notification**. Lampi AI notifies affected Customers promptly and without undue delay after becoming aware of a confirmed security incident involving Customer Data, and within 24 hours after awareness where contractually committed or required by applicable law for qualifying data breaches. Notifications may be provided through support channels, account contacts, contractual notice mechanisms, or , as appropriate. **Incident communications.** Incident communications may include the nature of the incident, affected systems or data categories, known or likely consequences, measures taken or proposed, mitigation steps, customer actions where relevant, and a Lampi AI contact point. Lampi AI may provide updates as investigation and remediation progress, and may issue closure communications or lessons learned where appropriate. **Availability monitoring**. Lampi AI monitors platform availability, application performance, system health, infrastructure behavior, network performance, and AI workflow execution. Alerts may be routed to internal communication channels and escalated according to severity. **Business continuity.** Lampi AI’s continuity practices include redundant and cloud-native infrastructure capabilities where applicable, backup procedures, monitoring, incident escalation, controlled deployment, and recovery planning. These measures are designed to reduce the likelihood and impact of service disruptions and support timely restoration of affected components. **Disaster recovery**. Lampi AI maintains disaster recovery procedures for restoring supported data stores, application components, configuration, and operational services. Recovery activities are coordinated by designated technical and security personnel, with executive oversight where required, and are supported by documented procedures and periodic testing. **Backup protection**. Backups are stored in Azure-hosted storage in the European Union where applicable and protected using secure transfer and encryption at rest. Backup access is restricted through least-privilege service credentials, infrastructure controls, and administrative safeguards. **Security review and improvement**. Lampi AI periodically reviews its security measures, supplier controls, risk posture, monitoring practices, incident procedures, and development processes. Improvements are prioritized based on risk, customer needs, regulatory expectations, operational maturity, and the evolving threat landscape for AI-enabled services.