A Practical Security Baseline for Business AI Installations
AI projects can touch websites, messages, documents, customer enquiries and third-party services. This page explains the delivery controls JASON AI uses as a baseline and the decisions that must be agreed for each project.
Start with the minimum system boundary
Production access should follow a defined task, not precede it. A project scope identifies the website, application, data source, user action and expected output involved. It should also identify what the installation must not access or change. This makes it possible to request the least privilege needed for the work and to remove or rotate access when the work is complete.
Where practical, changes are prepared in a staging area or copied to a controlled working directory. Live files are backed up before editing, and elevation is used only for the exact file or command that requires it. A service restart is not treated as a routine step and should be approved when production availability could be affected.
Controls expected in a managed installation
- Named systems, files and routes are inspected before a production change is proposed.
- Live-file backups include a timestamp and enough context to identify the related change.
- Credentials and tokens are not printed into reports, source control, chat transcripts or public pages.
- Administrative, account, payment and internal-tool pages are excluded from search discovery.
- Successful, error, mobile and accessibility paths are checked after deployment.
- Third-party model, hosting and integration dependencies are recorded in the handover.
Collect only what the workflow needs
An assistant or receptionist should not request personal information simply because a form can store it. Required fields are tied to a stated purpose such as returning an enquiry, preparing a quotation or checking whether a service fits. Sensitive or regulated information requires a separate decision about necessity, access, retention, deletion and qualified oversight.
Prompts, transcripts and uploaded documents may pass through third-party AI or infrastructure providers depending on the chosen architecture. The project scope must identify those providers and the customer must approve the intended data flow before live information is connected. JASON AI does not treat a general AI tool as an appropriate destination for confidential data by default.
From proposed change to verified handover
- 01Record the current state
Capture the relevant files, versions, routes, configuration and observable behaviour without exposing secrets.
- 02Back up exact targets
Create recoverable copies of the material files before a live edit or replacement.
- 03Validate the staged result
Run syntax, status, metadata, accessibility, mobile and security checks in proportion to the change.
- 04Deploy and verify again
Confirm the public response and document remaining limitations, monitoring and rollback information.
What this baseline does—and does not—mean
This page describes a delivery approach, not a certification, warranty of perfect security or substitute for a formal assessment. Risk depends on the system, data, jurisdiction, users, hosting and integrations chosen for a project. Customers remain responsible for lawful data collection, staff permissions, account ownership, appropriate notices, payment and tax obligations, and specialist review where their sector requires it.
Security also continues after launch. Software updates, credential rotation, access reviews, backup tests, logs, abuse handling and capacity monitoring need an owner and schedule. If ongoing management is included, those activities should be named in the service agreement rather than assumed.
Ask for a scope that names the data flow
Describe the current system, the information involved and the action the installation should take. JASON AI can then identify access, provider and testing requirements before production credentials are requested.