Security by design. Built in, not bolted on.
Every platform we deliver ships with the same controls we would demand as a client — encryption, private networking, least-privilege access, full audit trails and rehearsed recovery. Governed to the same standard whether it is a data pipeline or an AI agent.
Twelve controls that ship with every engagement.
Designed into the architecture from day one — not added at the end.
Encryption
TLS 1.2+ in transit, AES-256 at rest.
Private endpoints
Databases are never exposed to the internet.
Managed identities
No shared credentials between services.
Environment separation
Dev, staging and production fully isolated.
Audit logging
Every access and change is tracked.
Secrets management
Vaulted keys — no hardcoded credentials.
MFA + SSO
Enforced MFA, SSO via your identity provider — no shared accounts, engineer access time-boxed and logged.
Role-based access (RBAC)
Least-privilege by default; permissions attach to roles, not individuals.
Backup & disaster recovery
Automated backups, contractual RPO/RTO — restores rehearsed, not assumed.
Code review & deploy discipline
Every change reviewed, automated checks — no direct production edits, ever.
Version control integrity
Every commit content-hashed, history distributed across every clone — nothing lost if a server fails.
Credential hashing
Never plain text — salted, slow hashing (bcrypt / scrypt / Argon2) makes stolen credential stores useless.
Senior specialists · Security-first engineering · Production results
Agentic AI, governed like production infrastructure.
Client data never trains a model, sensitive fields are stripped before inference, retrieval respects permissions and anything that acts is human-released.
No training on client data
Contractual DPA terms, not policy.
PII redaction pre-inference
Sensitive fields stripped before any model call.
Permission-trimmed retrieval
The model never sees what the user can't access.
Action gating
Reading is free; acting is human-released.
Prompt injection defences
Input sanitisation and output validation.
DLP against shadow AI
Blocks data leaving to public AI tools.
Model evaluation before deployment
Benchmarked on your own documents and questions during pilot.
Confidence-based human review
Low-confidence outputs route to a human queue, not presented as certain.
Grounded answers with citations
Answers only from retrieved source material — every answer independently checkable.
Model-agnostic architecture
Not locked to one vendor — switching providers is configuration, not a rebuild.
Same standards as our core data platform · Nothing bolted on
Your data. Your rules, in every jurisdiction.
United Kingdom
UK GDPR compliant
You are controller, we are processor.
Data Protection Act 2018
Processing follows the ICO's guidance and principles.
NDA and DPA by default
Every engagement starts protected.
Saudi Arabia
PDPL aligned
Personal Data Protection Law, enforced by SDAIA.
NCA ECC aligned
Essential Cybersecurity Controls.
In-Kingdom hosting available
AWS Riyadh · Google Cloud Dammam · Oracle Riyadh / Jeddah.
Incident response & breach notification
Defined detect → contain → notify → remediate process, with contractual notification timelines — applies across both jurisdictions.
Professional indemnity + cyber insurance
In force, not pending. Certificates available to clients on request.
You own the data, always.
Retention, deletion and residency defined per contract.
Not yet ISO 27001-certified (small senior team) — our controls are built on its practices, and we're happy to complete your vendor security questionnaire.
Protected on paper. Owned by you at the end.
Contractual framework
Every engagement is backed by a standard set of agreements:
- Mutual NDA before any sensitive information is shared.
- Data Processing Agreement with a subprocessor annex.
- Statement of Work with defined acceptance criteria, RPO/RTO targets and support tiers.
- Support SLA with clear severity definitions and response targets.
AI subprocessors
Where an engagement includes AI, we use providers under enterprise data-use terms we have reviewed:
- Anthropic (Claude) — commercial terms with no training on customer data.
- Microsoft Azure OpenAI Service — data processed within the Azure tenancy, not used to train models.
- AWS Bedrock — model inputs and outputs are not shared with model providers or used for training.
Providers, regions and configurations for your engagement are documented in your DPA and its subprocessor annex.
Handover, not lock-in
Every environment is provisioned from a documented, repeatable infrastructure-as-code baseline. At the end of the engagement you receive full ownership of the code, infrastructure definitions, documentation, runbooks and administrative access. We design for handover from day one.
Open source & licensing
The legal layer under the stack.
Licences checked before use
Every open-source dependency is screened, favouring permissive licences (MIT, Apache 2.0, BSD) and flagging copyleft or source-available terms.
Dependencies pinned and scanned
Versions are locked and scanned for known vulnerabilities in CI, so an upstream change can't silently alter your system.
A clear bill of materials
At handover you get a list of every component and its licence, so your legal team knows exactly what you're running.
Need our policies, insurance certificates or a completed vendor security questionnaire? Get in touch or email info@xlr8tech.org.
Working systems, not slide decks.