Responsible use and organisational boundaries

Running models locally does not automatically:

• Bypass your employer’s data rules (you can still paste secrets into a local prompt).
• Prove a weight file is safe (malware can hide in bad downloads).
• Replace software approval processes (IT may still block installs).

This lesson is short guardrails plus doing: you will produce three IT questions you could really send.

Do this — trusted sources checklist (5 minutes)

Tick what you will do before installing any local runner or weights:

• Download only from official sites or package managers you trust.
• Read licence and privacy pages for the runner (Ollama, LM Studio, etc.).
• Prefer checksums or signed packages when the vendor publishes them.
• If something asks for admin access for no clear reason, pause.

Do this — three questions for IT (10 minutes)

Write three questions you would email IT before installing local inference on a work-issued laptop. Use your own words. Example shapes (do not copy all three verbatim):

1. “Are local LLM runtimes approved on our devices?”
2. “May model weights be stored on disk, and where must they live?”
3. “Do we require vendor support or security review for this package?”

Your three questions:

1. ---

2. ---

3. ---

Concept — local ≠ “no rules”

Licensing: open-weight models come with terms—commercial use, attribution, redistribution. Read them.
Employer data: “local” does not mean you may paste customer data into a hobby model.
Personal cloud: signing into your Google/Microsoft account on a work machine may still be a policy issue—same as Module 3.

Capstone bridge (one sentence)

Your capstone asks you to justify hosting and tooling in plain language—this module is where you practise justifying choices without jargon.

Check

• You can explain one malware risk of downloading weights from random links.
• Your three IT questions sound respectful and specific.

Key takeaways

• Trust the source; read the licence; ask IT on work hardware.
• Local is powerful, not permissionless.
• Module 05 returns to reading code—where fundamentals meet AI output.

What’s next

Next up: Module 04 check-in