AI security and compliance review: a risk officer reading a printed report at her desk

AI security and compliance

AI your security and compliance teams can approve.

How we handle your data in every solution we build: where it sits, who can reach it, what is recorded and where a person stays in charge.

Designed for the rules you answer to
  • Data residency
  • Encryption
  • Access control
  • Audit logs
  • Human oversight
  • Data retention
  • Singapore PDPA
  • MAS TRM Guidelines

Principles

Built on three commitments to the people who carry the risk.

They apply to every solution, in the cloud or on your own infrastructure.

  1. 01

    Your data stays yours

    It sits where you decide, it is kept out of model training, and it is deleted on the schedule you set.

  2. 02

    A person stays in charge

    The AI prepares and recommends. Decisions that carry risk are approved by the staff you name.

  3. 03

    Everything is recorded

    Each document read, question asked, answer given and action taken is logged with who, what and when.

Controls

The controls in every deployment.

Agreed with your security team during discovery and written into the design before the pilot starts.

Data residency

You choose the country and environment where documents, indexes and logs are stored and processed.

  • A named cloud region, or your own data centre
  • No transfer outside it without your approval
  • On-premise deployments can run with no internet
Deployment options

Encryption

Data is encrypted while it moves between systems and while it is stored.

  • TLS for data in transit
  • Encryption at rest for files, indexes and logs
  • Keys held in your own key store where you require it

Access control

People and systems see only what their role allows, and the AI works within the same limits.

  • Role-based access, linked to your sign-on
  • Restricted documents filtered out before the AI reads
  • Least-privilege accounts for every integration

Audit logs

A complete record of what the system read, decided and did, kept for your auditors.

  • Who asked, what was returned, which sources were used
  • Every approval and correction recorded
  • Exported to your logging or SIEM tools

Human oversight

Review steps are part of the workflow, set by you for each kind of decision.

  • Exceptions and low-confidence cases go to a person
  • Approval limits by amount or risk
  • Any automated step can be switched to manual

Data retention

Data is kept only as long as the purpose requires, then removed.

  • Retention periods set per data type
  • Deletion on request and at contract end
  • Sample data from pilots returned or destroyed

AI-specific safeguards

Protection against the risks that are new with AI.

Language models bring their own failure modes. These measures address them directly.

Guard models

A separate model screens requests and replies for attempts to extract personal data or steer the AI off its task.

  • Requests for other people's data blocked
  • Manipulation attempts flagged
  • Blocked requests logged for review

Grounded answers

The AI answers from your documents and data, and shows where each answer came from.

  • Sources cited on every answer
  • A clear reply when the documents do not cover a question
  • Rule checks shown with their working

Testing before release

Each system is tested on a set of your real cases, and tested again whenever a model or rule changes.

  • Accuracy measured on your own samples
  • Results shared with your team
  • Changes released only after the tests pass

Frameworks

Designed for the frameworks you are assessed against.

We map the design to the obligations that apply to you and document it for your assessors.

Singapore PDPA

Solutions are designed for the Personal Data Protection Act's obligations on consent, purpose, protection, retention and transfer.

  • Personal data identified and minimised
  • Purpose and retention documented
  • Support for access and correction requests

MAS Technology Risk Management

For financial institutions, we design for the MAS TRM Guidelines and support your outsourcing and third-party risk assessments.

  • Documented architecture and data flows
  • Change, access and incident procedures
  • Evidence prepared for your risk team

Your own policies

Your security standards and vendor requirements are treated as part of the specification.

  • Vendor security questionnaires completed
  • Your penetration test findings addressed
  • Deployed inside your environment and controls

Have a security questionnaire?

Send it over. We'll answer it in full and walk your team through the architecture.

Book a call

FAQ

Security questions, answered.

Something else on your mind?

Ask us directly
AI security review meeting: two people discussing a checklist in a glass-walled meeting room

Security review

Bring your security team into the first conversation.

We'll walk through the architecture, the data flows and the controls for your use case, and answer your questionnaire.

Book a call
  • Architecture and data flows explained
  • Questionnaires answered in full
  • Private deployment available