06 / AI Explained · Field briefing
Your data has a route
Follow information beyond the model.
“Local” and “cloud” are useful starting labels, but neither describes an entire information journey. A local application can call a remote model. A cloud model can receive only redacted text while originals stay elsewhere. A local model can still expose information through logs, backups, plugins or exports.
Where does information go when I use AI?
An explanatory schematic—not a live system or measured performance.
KAF World Digital · Prepared 4 October 2026 · First editorial edition. Worked examples are invented for teaching.
Try the example02 / Inspect every crossing
Inspect every crossing
Draw the route for one task. Identify the input, preprocessing, model endpoint, tools, output store, logs and backup locations. Then identify who controls each step. This makes questions about access, retention and deletion concrete enough to investigate.
An invented document-summary workflow illustrates the distinction. A team could send a full customer file to an approved remote service; it could remove direct identifiers before sending; or it could run an appropriate model on controlled local hardware. Each option has different quality, operational and data-handling requirements. Redaction reduces some disclosures but is not automatically anonymisation: a document's contents may identify a person without a name.
Check policies and settings for the exact service and plan. Questions include whether inputs are used for training, which logs are retained, where processing occurs, who can access content, which subprocessors participate and how deletion works. A statement about one product should not be silently applied to another product from the same vendor.
For local operation, ask who administers the machine, where model and application files are obtained, how updates are reviewed, what network access exists and where backups land. Owning the hardware changes some boundaries; it does not remove operational responsibilities.
For hybrid systems, list the crossings explicitly. A document's original text, embeddings, snippets and generated summaries may travel separately. Embeddings are derived representations, not proof that the source information has become non-sensitive.
Start with the minimum authorised information needed for the task. Keep access and secrets outside prompts where possible, and test with synthetic data while the route is being designed. Legal and regulatory obligations depend on the actual processing context; this guide is a decision framework, not legal advice or a claim of compliance.
Evidence for the technical mechanisms: Microsoft · Using Generative AI Responsibly, Microsoft · Introduction to Small Language Models. The worked examples and practical framework are KAF editorial recommendations.
03 / Worked example
Compare three paths
Local, cloud and hybrid are starting points—not privacy guarantees. This interactive lesson uses prepared synthetic examples and local calculations.
Use the companion worksheet and the worked example. All assumptions are illustrative.
No account, model API or external service is connected. Inputs are not submitted or stored by this page.
04 / Go deeper
Useful words. Important details.
- Data boundary
- A crossing between systems, owners or access scopes.
- Retention
- How long information or derived copies remain stored.
- Derived representation
- A transformed version such as an embedding or summary that can still reveal information.
- Anonymisation
- A context-dependent standard of preventing identification; removing a name is not enough by itself.
Map derived copies
Inventory snippets, embeddings, summaries, logs and caches as well as original files. A transformed representation can retain sensitive information, so a claim about the source location is not a complete processing map.
Verify the exact service
Retention, training use, regional processing and deletion differ by product, plan and configuration. This briefing provides questions rather than unsupported provider assurances.
Minimise the task input
Supply the minimum authorised information required. Check whether combined innocuous fields identify someone or reveal a confidential event. A local machine still needs access controls and a backup policy.
The technical notes qualify the explanation; they are not product-specific assurance or a certification.
05 / Evidence & scope
Inspect the source.
Keep the context.
Sources were accessed on 4 October 2026. These technical and security references support the mechanisms described; they do not validate our fictional examples or establish comparative product performance.
- Microsoft · Using Generative AI Responsibly
Responsible-use principles, representative testing and operational ownership. Technical guidance, not legal advice.
- Microsoft · Introduction to Small Language Models
Deployment considerations for smaller models. No specific hardware or privacy guarantee is inferred.
- OWASP · Sensitive Information Disclosure, 2026
Information can leak through retrieval, logs, derived representations and outputs. This briefing uses the general risk framing, not legal conclusions or incident statistics.
- Hugging Face · Secure code execution
Code-execution risks and layered execution controls; no universal safety guarantee.
What this briefing does—and does not—establish
The worked examples, suggested questions and worksheets are original KAF teaching material. Numbers are explicitly illustrative. No customer data, observed adoption results, market-size forecast or live provider pricing is presented.
Check the exact product, release, configuration and policy before using a workflow. Source documents can change; this is a dated educational briefing.
Keep a useful copy.
Read the full briefing offline, or use the companion worksheet to apply it.