Localign
Controls

84 controls, in five domains

The technical and organisational controls behind the product: infrastructure, organisation, product, AI safety, and data and privacy. Filter by domain, or request the full security dossier.

Download documentsView the 84 controls
Questions about your assessment?trust@localign.com
All systems operationalService status
HostingEU (Netherlands, France)
ModelsOpen source, EU-hosted
Training on customer dataContractually prohibited
Escalation to an external modelConsent per question
Quick links
ISO 27001Sub-processor registerDPA + AI AnnexRetention scheduleFair use policyService status
Domain

Infrastructure security

12 controls
Hosting and Mode A processing within the EEA; Mode B providers may process outside the EEA, only after personal-data screening; every sub-processor is documented with its location
Encryption at rest (AES-256) and in transit: TLS 1.2 and 1.3 only, with forward-secret ciphers; HTTP is redirected to HTTPS, and HSTS is set with preload
Segregated development and production environments: separate hosts, separate databases and separate deployment runners, with changes reaching production only through the protected main branch
Key management with rotation and separated access
Backups with tested restore procedure and defined retention
DDoS protection by our hosting providers (Scaleway and TransIP) and rate limiting at the edge
Patch and vulnerability management with SLA per severity
Business continuity and disaster recovery plan, tested annually
Logical separation of customer data (multi-tenant isolation), enforced server-side on every request and tested across organisations
Only HTTP and HTTPS are exposed to the internet; all other ports are closed, and internal and administrative services are reachable only through the company VPN. Certificates are managed separately from the application runtime
Deployment switching at the public ingress layer to reduce cut-over risk during releases
Email domains protected against spoofing with a DMARC reject policy and SPF

Organisational security

16 controls
Information Security Management System (ISMS) with scope, policy and risk register
Least-privilege and need-to-know access policy
Onboarding and offboarding procedure with access review
Documented roles and access levels for employees and contractors with access to customer data
Confidentiality agreements for employees and contractors
Security awareness training, annual and at hire
AI literacy training (mandatory under AI Act art. 4)
Incident response procedure with roles, escalation paths and drills
Detached conversation flows: a user message keeps processing on the server even if the user's tab disconnects, and the resulting state is restored when they return, with no silent data loss on browser refresh
Documented role ownership for security, privacy and incident-response responsibilities
Short-lived, per-user tokens for internal document retrieval
Vendor management with DPA, security review and annual reassessment
Risk assessment process, annual and on material change
Privacy point of contact for data subject and regulator requests
Internal audit programme
Management review, at least annually

Product security

19 controls
Secure SDLC with code review, SonarQube static analysis and dependency management
SSO authentication with mandatory MFA for administrators
Organisation-level switch to disable Google and Microsoft social login while keeping password and Keycloak login paths available
Role and group-based authorisation
Audit visibility for model use, sources and PII consent events inside the product
Audit log of sensitive data shared with external AI models
Data validation and input sanitisation
Penetration test, annual and on major releases
Rate limiting on public invitation and auth entry points, plus WebSocket abuse controls keyed to trusted-proxy client identity
Version control and release notes
Secrets management (no credentials in code)
API rate limiting and abuse protection
Short-lived access JWT in the browser (5 minutes), with a token version for server-side revocation; the refresh token is kept in an HttpOnly cookie rather than readable client-side storage
Session refresh before protected HTTP requests and WebSocket reconnects, with fail-closed behaviour after repeated auth rejection
Per-answer inspection of models used, document and web sources, and PII audit events
Strict token validation on the API and the WebSocket: unsigned, tampered or forged tokens and tokens with a swapped algorithm are rejected
Unauthenticated API requests are refused, and the API documentation is not publicly exposed
No secrets, API keys or source maps in the code delivered to the browser
Assessment of the external attack surface, most recently in October 2026

AI safety

25 controls
Documented risk classification of the AI system
Intended purpose and excluded use per Annex III category
Model inventory with versions, providers and intended uses
The organisation administrator decides which external AI providers are allowed and can enable or disable each provider; a new external provider is never enabled for an existing organisation without its explicit consent
Sensitive-data detection runs before any transfer to an external model: under Mode B the request is blocked when unconsented personal data is found
Layered sensitive-data detection: deterministic regex and checksum, parallel specialist detectors, and a contextual-judge verifier
Input normalization before detection: Unicode and irregular whitespace are collapsed so simple obfuscation tricks do not bypass the PII scan
Consent dialog when a question requires sensitive data; optionally enabled for every external question
Per-provider consent state: granting consent for one external provider does not grant it for any other
Unavailable or disabled providers are blocked explicitly, never silently substituted with another provider
External queries are deduplicated and capped to a small number of provider calls per turn
Customer attestation at signing: no article 5 prohibited practices, with operational enforcement
Configurable reasoning effort: customers choose how much thinking-time models spend per question, balancing cost and depth
URL-context guardrail: pasted URLs only reach the model that has the URL-context tool enabled, never silently to other providers
No silent fallback: if the primary model fails, fallback chains stay within the same provider family or its declared substitutes, never out of region
Conversation history, raw documents and raw images do not travel to external providers; only the filtered question text can be sent externally
Models are evaluated continuously for answer quality on fixed test sets, such as the Dutch final exams and IBM's MTRAG retrieval benchmark, and for bias; results are published as insights on this website
Use-case guardrails with blocklist for prohibited practices (AI Act art. 5)
Transparency to end users (AI Act art. 50)
Human oversight capability (AI Act art. 14)
Fundamental Rights Impact Assessment (AI Act art. 27) building blocks published for deployer use
Post-market monitoring with feedback mechanism
AI incidents are handled under our incident procedure. Article 73 of the AI Act applies to high-risk AI systems; Localign is not used as one unless a customer agrees an Annex III use with us first
Documented model training policy (no training on customer data)
Suppression or labelling of AI-generated output where relevant

Data and privacy

17 controls
Records of processing activities (GDPR art. 30)
Data location overview per data category
Retention periods defined per type of data; your data is kept for as long as your contract runs; backups and previous document versions expire automatically
After termination, 30 days to export your data (extendable by up to 60 days on reasoned request); then deletion from active systems within 30 days and from backups within 35 days after that, confirmed in writing within 10 working days
Data deletion procedure with confirmation to the customer
Data subject request procedure (GDPR art. 15-22)
Breach notification procedure (within 48 hours to the customer)
DPIA template and completed DPIAs for high-risk processing
Privacy by design and privacy by default in product development
Data minimisation and pseudonymisation where possible
International transfer mechanisms documented (SCCs for Mode B)
Sub-processor management with 30-day customer notification
Controller and Processor role split per use case
Document access scoped by ownership, organisation membership, assistant sharing and project rights before retrieval is allowed
Attached-document and assistant-library identifiers passed as explicit allow-lists to the internal document path
Deleting a document removes its chunks, images and summaries at once; backups and previous versions of the file expire within 30 days
Presigned document download URLs expire automatically after one hour

Another question about your assessment?

We usually answer questions from CISOs, DPOs and procurement teams within two working days.

trust@localign.comBook a demo