Private client environment
Authoritative and identifiable information
- Original customer records
- Customer and supplier names
- Product numbers and contact details
- Confidential specifications
- Commercial information
- Final approved responses
Security, Confidentiality and Data Handling
Evidence-backed work requires more than a confidentiality promise. It requires separation, minimum access, human verification and visible client authority.
Read the AI boundaryCore operating rule
“No original customer document, confidential product file or identifiable customer information is submitted to a public or personal AI account. AI-assisted work is limited to fictional, anonymised or appropriately pseudonymised working information, following documented client instructions. All final outputs are checked by an authorised human against the original evidence before submission.”
A. Customer information
Access follows the agreed service.
Information is processed only for the documented scope and is limited to authorised personnel.
A project may require controlled access to the following types of customer information:
- Product questionnaires
- Technical data sheets
- Bills of materials
- Supplier declarations
- Certificates and product manuals
- Customer evidence requests
- ERP or spreadsheet exports
The public website does not provide a general document-upload function. Evidence is requested only after qualification, confidentiality discussion and written agreement.
B. Separated environments
Original records stay separate from AI-assisted drafting.
The dividing line applies to documents, names, identifiers and commercially sensitive context—not only to personal data.
AI-assisted working environment
Minimum, controlled working information
- Fictional demonstration information
- Redacted questions
- Reference-coded products
- Extracted non-identifying facts
- Pseudonymised evidence summaries
- Draft structures and wording
Pseudonymised does not mean anonymous. It may still be personal or confidential information and must continue to be protected, access-controlled and handled under the agreed instructions.
C. Human review
AI output is never evidence.
Every answer must be supported by client-provided or client-authorised records. Unsupported claims are marked as gaps, conflicting evidence is escalated and the client confirms accuracy and disclosure.
D. Data minimisation
Request the smallest useful set.
Only information reasonably required for the agreed work should be requested, accessed or transferred. Irrelevant identifiers and document sections should be excluded where practicable.
E. Retention and deletion
The contract sets the instruction.
Project retention periods are agreed in the client contract. Documents are returned, retained or securely deleted according to those instructions and applicable legal requirements. No universal period is assumed.
F. Service providers
Third parties require controls.
External providers are used only where appropriate contractual, confidentiality and security controls are in place. Named subprocessors will be published only after the production technology stack is approved.
G. Designed security controls
Control intent—without pretending to hold a certification.
Our production system is designed to use the controls below. Final implementation, supplier due diligence and testing must be confirmed before a secure client portal is released.
Administrator and authorised-user access
Only the permissions required for the agreed work
Named recipients and project-specific access
Protected transfer between approved systems
Client-approved production repositories
Access, version and release events where supported
Documented recovery appropriate to the service
Escalation, containment and notification procedures
Revocation after completion or role change
Secure client portal — production implementation pending.
No visual login page is presented as a working secure system. Authentication, recovery, session control, audit logging and vulnerability testing must be complete before release.
Production release gate
Required before a website backend or client account area handles evidence.
- HTTPS across the whole website
- Secure session management
- Server-side validation and input sanitisation
- Rate limiting and anti-spam protection
- Content Security Policy and secure HTTP headers
- No credentials stored in website code
- Environment variables or managed secrets
- Encrypted database connections
- Role-based client access
- Multifactor authentication for administrators
- Audit logging
- Secure password-reset process
- Regular dependency updates
- Backup and recovery procedures
- Document-access expiry
- Account and access revocation
- Independent vulnerability testing before launch
Related controls
Read the operating boundaries together.
Start with the live request
Turn the live request into a controlled response pack.
Check fit privately, then start a non-confidential conversation about the deadline, scope and evidence condition. No files are uploaded through this website.
Check pilot fit Email a general enquiryDownload the pilot brief Do not attach confidential questionnaires or evidence until a secure sharing route is agreed.