Chapter 28 — Field practice | Technologist AI and machine-learning security

Open, free learning for anyone who wants to protect people and technology. No account is required to read this text or its worked answers.

Chapter 28 — Field practice

Lesson map

Lesson 28 published. Map, example, EX082.

flowchart TD
  N0["Learning outcomes"]
  N1["28.1 Placement scope"]
  N2["28.2 Typical work"]
  N3["28.3 Practice log"]
  N4["28.4 Reflection"]
  N5["Synthesis"]
  N0 --> N1 --> N2 --> N3 --> N4 --> N5
  N5 --> EX["Worked example"]
  EX --> Q["Exercise EX082"]

Logical drawing

flowchart LR
  N0["Learning outcomes"]
  N1["28.1 Placement scope"]
  N2["28.2 Typical work"]
  N3["28.3 Practice log"]
  N4["28.4 Reflection"]
  N5["Synthesis"]
  N0 --> N1 --> N2 --> N3 --> N4 --> N5 --> Q["EX082"]

[Learning outcomes]
    |
    v
[28.1 Placement scope]
    |
    v
[28.2 Typical work]
    |
    v
[28.3 Practice log]
    |
    v
[28.4 Reflection]
    |
    v
[Synthesis]
    |
    v
[Worked example] --> [Exercise EX082]

Learning outcomes

You will (1) apply AI-security skills during a supervised placement; (2) work within an organisation's policies and authorisation; (3) keep a practice log; and (4) reflect on gaps between course and practice.

28.1 Placement scope

A placement agreement states your supervisor, the systems you may work on, what you may test, data rules and reporting lines. Read the organisation's acceptable-use and AI policies before starting. If something you are asked to do is outside the agreement, ask before doing it; a polite question is always cheaper than an unauthorised test. Jonas's placement agreement with customer-service operations named his supervisor, the systems he could review and the data he could see. When he noticed agents pasting customer emails into a public chatbot, he did not test the chatbot or collect examples; he raised it with his supervisor, which was within the agreement.

28.2 Typical work

Placements often involve: updating the AI inventory; reviewing a vendor AI feature; adding cases to an evaluation suite; writing a runbook; investigating an alert with a senior colleague. Small, finished tasks build trust faster than ambitious unfinished ones, and they leave something useful behind when you go. Jonas's placement produced a finished task of this kind: a one-page guide on using the approved internal assistant instead of a public chatbot. It said which customer data may go into which tool and how agents request access to the approved assistant.

28.3 Practice log

Record each week: tasks, decisions, evidence produced, what you learnt, questions raised. Logs support assessment and become a portfolio. Keep confidential details out — describe work in general terms where needed, and ask your supervisor what may be included. Jonas's log for the shadow-AI week described the finding in general terms, without names of agents or quotes from customer emails. His supervisor checked the entry before it went into his portfolio. The log also recorded the question he raised and the decision the team made.

28.4 Reflection

Practice differs from coursework: systems are messier, documentation is missing, priorities change. Reflect on which course methods worked unchanged, which needed adapting, and what you still need to learn. Jonas's reflection noted that the inventory method worked unchanged, but finding shadow use required talking to people. That is a lesson coursework had not given him, and it shapes how he would approach a new inventory. He also noted that a scan would have found the approved assistant but never the public chatbot, because nothing in Brightwell's systems recorded its use.

28.5 Worked example (synthetic Brightwell case)

Jonas's placement is with Brightwell's customer-service operations team. He discovers agents pasting customer emails into a public chatbot to draft replies — shadow AI with personal data. Rather than testing or reporting individuals, he raises it with his supervisor; the team adopts the approved internal assistant and Jonas writes a one-page guide. His log notes that the inventory method from Chapter 11 worked, but finding shadow use required talking to people rather than scanning systems. Treating the problem as a gap in tools and guidance, not misconduct, made agents more willing to switch.

28.6 Errors and limits

Acting outside the placement agreement.
Confidential details in logs.
Blaming individuals for systemic problems.

Jonas's placement avoided these. He stayed within the agreement, kept customer details out of his log, and treated shadow AI as a gap in tools and guidance rather than individual blame. Reporting individuals would likely have driven the practice underground rather than ending it, and the approved assistant solved the need agents had. The approved assistant keeps conversations inside Brightwell's own systems, which is the part of the fix that protects the customer emails. The guide named no individuals.

Exercises

EX082 — Agreement. List six elements of a placement agreement. Rubric: 3 = six apt; 2 = four or five; 1 = two or three; 0 = fewer.
EX083 — Shadow AI. Explain how Jonas handled the shadow AI and why that approach is effective. Rubric: 3 = escalation, systemic fix, guidance, no blame; 2 = two; 1 = one; 0 = none.
EX084 — Log entry. Write one week's practice-log entry. Rubric: 3 = tasks, decisions, evidence, learning, questions, no confidential detail; 2 = four; 1 = two; 0 = none.

Synthesis

Field practice applies course methods under real constraints, within authorisation, and turns experience into reflection and a portfolio.

Continue studying — chapter 28 of 30

← Chapter 27 (draft) · Course chapter index · Chapter 29 (draft) →