Lesson 6

Your organization’s AI rules, and when to ask

Video
4 min
Reading
14 min
Exercise
12 min
Total
29 min

Educational content, not legal advice.

In this lesson you will

  1. Find your way around your organization’s AI acceptable use policy and know that it takes priority.
  2. Recognize the situations where you ask before you act, and whom to ask.
  3. Report an AI incident quickly with the right facts, and explain why speed matters.
  4. Explain in plain words why the law cares about AI literacy in Canada and the EU, and keep your own AI literacy record.

Video lesson

Length: 3:04
Read the transcript

Welcome to the last lesson of AI Literacy at Work. Lesson six: your organization’s AI rules, when to ask, and why the law cares.

Most organizations put their AI rules in one place, often called an acceptable use policy. It lists the approved tools, and what information can go into each. It says who checks output, and when to say AI was used. And it explains how to ask for a new tool, and how to report a problem. Read your own. Where it differs from this course, it wins.

So when should you ask before you act? Ask when a tool or feature is new to you. Ask when the information is new, or sensitive. Ask when AI would shape a decision about a person, or when the result will be public. And ask when you cannot check the output, or something simply feels off. Ask your manager, the privacy officer, IT, or legal, depending on the question. Asking before costs two minutes. Fixing after can cost weeks.

If something does go wrong, report it right away. Write down the facts: which tool, which account, what information, and when. Then your organization decides what happens next. Speed matters. Where PIPEDA applies, a breach of security safeguards that creates a real risk of significant harm is reported to the Privacy Commissioner as soon as feasible, and the affected individuals are notified. PIPEDA s.10.1(1)-(3), (6) In Quebec, a confidentiality incident that presents a risk of serious injury is reported promptly to the Commission d’accès à l’information and to the people concerned. P-39.1 s.3.5 Your organization cannot do any of that until it knows.

Why does the law care about AI literacy? In Canada, there is no federal AI statute in force. Bill C-27 Part 3 (44-1) But where privacy law applies, it reaches AI use. Where PIPEDA applies, personal information has to be protected by safeguards that suit how sensitive it is, whatever tool handles it. PIPEDA Sch.1 cl. 4.7

In the European Union, the AI Act goes further. Providers and deployers of AI take measures to support the AI literacy of the staff and others who operate or use AI on their behalf. AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) The Act sets no specific level. AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) This duty has applied since February second, twenty twenty five. Regulation (EU) 2024/1689 Art. 2(2), Art. 111(4), Art. 113, as amended by Regulation (EU) 2026/1744

That is why many employers keep a record: who took which training, on which tools, and when. At our fictional property manager, Quillfield, every employee signs an AI literacy record after this course, and updates it when they start using a new tool. Keep yours current when your job or your tools change.

Recap: know your rules, ask before you act, report fast, and keep your record up to date. Thank you for completing AI Literacy at Work. Take the final exam to earn your certificate.

Notes

Your organization’s rules come first

This course gives you general habits. Your organization’s own rules apply to your actual job. Most organizations put them in one document, often called an acceptable use policy for AI, sometimes inside a wider IT or privacy policy. Where it differs from this course, your organization’s policy wins.

A typical policy covers these points. Use the list to find your way around your own.

SectionWhat it tells youWhere this course covers it
Approved toolsWhich tools you may use for work, on which accountsLessons 1 and 3
Information rulesWhat can go into which tool, and the never listLesson 3
Checking and authorshipWho checks AI output, and how carefullyLesson 2
DisclosureWhen to say AI was used, inside and outsideLesson 4
Decisions about peopleWhere a person reviews AI output, and howLesson 4
SecurityAccounts, connected apps, verification of requestsLesson 5
New toolsHow to ask for a tool or feature to be reviewedThis lesson
IncidentsWhat to report, to whom, and how fastThis lesson
TrainingWhich training you take, and the record keptThis lesson

If you cannot find your organization’s policy, ask your manager. If there is none yet, use this course’s habits and the templates, and suggest that one be written.

When to ask first

Asking costs two minutes. Fixing a mistake that reached a customer, a tenant or a regulator can cost weeks. Ask before you act when:

  • The tool or feature is new to you, including a new AI button in software you already use.
  • The information is new or sensitive: personal information you have not used with AI before, anything on the never list, confidential documents.
  • AI would shape a decision about a person: hiring, renting, discipline, collections, eligibility.
  • The result will be public: social media, a website, an ad, a notice to many people.
  • The topic is law, money, health or safety, and you would rely on what the tool says.
  • You cannot check the output because you do not have the source or the knowledge.
  • Something feels off: an odd request, a tool that acts on its own, an answer that seems too good.
AskAbout
Your managerWhether a use fits your job and team, and anything you are unsure of
The privacy officerPersonal information, new uses of data, a paste mistake
IT or securityNew tools, connected apps, suspicious messages, account problems
Legal or complianceAnything about what the law says, contracts, intellectual property

Asking for a new tool

Good tools get approved faster when the request is clear. Say what the tool is, what task it would help with, what information would go into it, who would use it, and what you would do if it got something wrong. Then wait for the answer before you use it for work.

Reporting incidents

Report an AI incident right away, even if it seems small, and even if you are not sure it counts. Typical AI incidents:

  • personal or confidential information put into the wrong tool;
  • a wrong AI output that reached a customer, a tenant or a decision;
  • an AI tool acting on its own: sending, deleting or changing something you did not ask for;
  • a suspicious message, call or video that may use AI to impersonate someone;
  • an outcome that looks unfair to a group of people.

Write down the facts: which tool, which account, what information, what happened, when, and what you have done so far. Do not try to fix it quietly. Your organization decides what happens next.

Why speed matters. Some incidents involve legal steps with their own timing, and your organization cannot take them until it knows.

  • Where PIPEDA applies, a breach of security safeguards that creates a real risk of significant harm is reported to the Office of the Privacy Commissioner of Canada as soon as feasible after the organization determines it has occurred, and the affected individuals are notified. PIPEDA s.10.1(1)-(3), (6)
  • In Quebec, if a confidentiality incident presents a risk of serious injury, the business notifies the Commission d’accès à l’information and the people concerned promptly. P-39.1 s.3.5
  • Where the GDPR applies, a personal data breach is notified to the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of it, unless it is unlikely to result in a risk. GDPR Art. 33(1)

You do not decide whether these apply. You make sure the people who decide hear about it quickly.

Why the law cares about AI literacy

In Canada, there is no federal AI statute in force. Bill C-27 Part 3 (44-1) That does not leave AI unregulated. In December 2023, Canada’s federal, provincial and territorial privacy authorities stated that generative AI does not sit outside current legislative frameworks, and that organizations developing, providing or using it must comply with applicable privacy laws. OPC and FPT privacy authorities, Principles for responsible, trustworthy and privacy-protective generative AI technologies (December 7, 2023) Where PIPEDA applies, personal information has to be protected by safeguards that suit how sensitive it is, whatever tool handles it. PIPEDA Sch.1 cl. 4.7

In the European Union, the AI Act addresses AI literacy directly.

A deployer is a person or body using an AI system under its authority, except in a personal, non professional activity. AI Act Art. 3(1), (3) and (4) So an organization whose staff use AI systems at work can be a deployer, and the literacy duty can reach it if the Act applies to it at all.

What this means for you in practice: your employer may ask you to take training like this course, to follow its AI rules, and to sign a short record of what you learned and which tools you use. That record is part of how an employer shows the measures it took.

Your AI literacy record

The template for this lesson is an employee AI literacy record. It is a one page record you and your manager fill in together: the training you took, the tools you are approved to use and for what, the rules you have read, and the date of your next refresher. Employers keep it as evidence of the measures they take. Keep it current when you start using a new tool or your job changes.

The course in six habits

  1. Treat every AI answer as a draft. You are the author.
  2. Check the source of every fact that matters, and match the check to the stakes.
  3. Never paste the never list. Use approved tools, and paste only what the task needs.
  4. Say when AI was used, and keep a real person in charge of decisions about people.
  5. Verify unusual requests on a known channel, and guard your accounts.
  6. Ask before you act, and report problems fast.
Worked example

Quillfield writes its rules on one page

Fictional organization. Quillfield Property Management Inc. is invented for this course. Any resemblance to a real company is a coincidence.

After the events in the earlier lessons (the rental application in a free chatbot, the applicant score nobody questioned, the cloned voice), Quillfield’s operations director asked the privacy officer and the IT manager for "rules people will actually read". They wrote one page.

Quillfield’s AI rules, in short.

  1. Approved tools: one chat assistant on the business account; the meeting notes feature for internal meetings only; the maintenance urgency sorting; the screening score as one input, never the decision. Anything else is unapproved until IT says yes.
  2. Never paste: passwords and codes, ID numbers, bank and card numbers, health information, full files about a person, legal documents.
  3. Check: every fact that matters has a source. Notices to many tenants get a second reader.
  4. Say so: tenants know when they chat with a bot. AI images never pass as real photos.
  5. People decisions: a person decides, with the file and the authority to disagree with the tool.
  6. Security: new payees and bank changes are confirmed by a call back and a second approver.
  7. Ask first: new tools, new data, decisions about people, anything public, anything about law, money, health or safety.
  8. Report fast: to the privacy officer or IT, the same day, using the incident form.
  9. Training: this course in the first month, a refresher every year, and a signed AI literacy record.

How the rules are kept alive.

  • The page is pinned in the staff chat and printed at each office.
  • Each new AI tool request goes to IT with five answers: what the tool is, the task, the information, the users, the plan if it goes wrong. Most get an answer within a week.
  • The privacy officer keeps the incident forms and reviews them every quarter with IT. Two patterns from the first year became new training examples.
  • Each employee’s AI literacy record lists the course date, the approved tools they use and what for, and their next refresher date. The operations director can show the full set if a client or regulator asks what Quillfield does to support AI literacy.

Why it works. The rules are short, specific to Quillfield’s tools and tied to real incidents. Staff know where to ask, and asking is quicker than guessing.

Scenario exercise · Allow 12 min

Ask, report, or go ahead?

For each situation, decide: go ahead, ask first (say whom), or report (say what you would write down). Assume your organization has an approved chat assistant and a policy like Quillfield’s.

  1. You want the approved assistant to tidy up the wording of an internal agenda for a team meeting.
  2. A colleague shows you a free AI app that turns tenant photos of damage into repair cost estimates, and suggests you both start using it.
  3. You notice that last week you used the approved assistant to summarize a tenant’s email, and the summary that went into her file says she "refused entry to the technician". Her email actually said she asked for a different day.
  4. Your manager asks you to use the approved assistant to draft a public social media post about a rent freeze at one building.
Show the model answer
  1. Go ahead. Approved tool, internal document, no personal information, low stakes. Read the result before you send it.
  2. Ask first: IT (new tool) and the privacy officer (tenant photos can show people and their homes). Do not use it while you wait. Cost estimates also feed money decisions, so the tool’s accuracy needs checking before anyone relies on it.
  3. Report, to your manager or the privacy officer, and fix the record through the proper process. Write down: the tool, the date, what the summary said, what the email said, where the summary was stored, and who may have relied on it. The wrong summary is inaccurate personal information about the tenant, and it could affect how she is treated.
  4. Go ahead with care, and ask the person who approves public posts. A public post about rent is high stakes: check every fact (which building, which dates, which units), have a second person read it, and follow your organization’s rule on public communications.

Why this is the strongest answer

The four cases show the range. Low stakes work in an approved tool needs no permission, only a read. A new tool always goes through a request. A past mistake that reached a person’s file is an incident, even though it is "only" a summary: accuracy of personal information matters, and the people responsible for the record need to know. Public content gets a second reader. The strongest answers say not just what to do but whom to involve, because knowing where to ask is the skill this lesson teaches.

Template to download

Employee AI literacy record

A one page record that an employee and their manager fill in and the employer keeps as evidence of its AI literacy measures: training taken, approved tools and uses, rules read, acknowledgements and the next refresher date.

Check your understanding

Scenario questions. Answer them all, then check: you will see why each option is right or wrong.

Question 1 of 5

You cannot find your organization’s AI policy, and your manager says there is none yet. What is the most sensible approach?

Question 2 of 5

Which situation most clearly calls for asking before you act?

Question 3 of 5

Where PIPEDA applies, a breach of security safeguards that creates a real risk of significant harm is reported to the federal privacy commissioner’s office. When?

Question 4 of 5

A colleague says: "The EU AI Act makes every employee reach a set level of AI skill, checked by an exam." Is that right?

Question 5 of 5

What should an employee AI literacy record contain to be useful as evidence?

Sources cited in this lesson

  1. PIPEDA s.10.1(1)-(3), (6) official text current to 2026-09-30, checked on 2026-11-03
  2. P-39.1 s.3.5 official text current to 2026-10-15, checked on 2026-11-01
  3. GDPR Art. 33(1) official text current to 2016-05-04, checked on 2026-11-01
  4. Bill C-27 Part 3 (44-1) checked on 2026-10-24
  5. OPC and FPT privacy authorities, Principles for responsible, trustworthy and privacy-protective generative AI technologies (December 7, 2023) official text current to 2025-05-06, checked on 2026-11-01
  6. PIPEDA Sch.1 cl. 4.7 official text current to 2026-09-30, checked on 2026-10-24
  7. AI Act Art. 4(1), as replaced by Regulation (EU) 2026/1744; Art. 113, third paragraph, point (a) official text current to 2026-07-27, checked on 2026-11-01
  8. Regulation (EU) 2024/1689 Art. 2(2), Art. 111(4), Art. 113, as amended by Regulation (EU) 2026/1744 official text current to 2026-07-27, checked on 2026-11-01
  9. AI Act Art. 3(1), (3) and (4) official text current to 2026-07-27, checked on 2026-10-29