10  10. Legal and Ethical Questions About AI

This page continues Limits, Risks, and Responsible Use. It covers copyright, who is responsible for AI output, who owns AI-generated code, academic integrity, regulation, and what AI means for jobs. The same idea runs throughout: you stay responsible for what you do with AI.

10.2 Who is responsible for AI output?

You are, and so is your organization. Courts have penalized lawyers for filing AI-invented citations, and in 2024 a Canadian tribunal held Air Canada responsible for wrong refund advice from its website chatbot, rejecting the idea that the chatbot was responsible for its own actions (CBC). If AI-written code leaks customer data, “the AI did it” is not a defense.

10.3 Who owns AI-generated code?

The US Copyright Office’s position is that purely AI-generated material is not protected by copyright, while human contributions (choosing, arranging, and modifying the output) can be (US Copyright Office). AI can also produce code that closely resembles existing open-source code, which comes with license conditions (such as keeping the original copyright notice). Check the licenses of libraries the AI adds, and follow your employer’s policy on AI-generated code.

10.4 Academic integrity

Each course sets its own rules, so read the policy for every assignment. Disclose AI use when required, never submit citations you haven’t checked, and be able to explain everything you hand in. Even where AI is a core tool, as in this course, you are responsible for the result.

10.5 Regulation

The EU AI Act regulates AI by level of risk. It also applies to non-EU companies whose AI is used in the EU (European Commission).

Date What applies
Aug 1, 2024 The law enters into force
Feb 2, 2025 Bans on “unacceptable-risk” uses, such as emotion recognition in the workplace
Aug 2, 2025 Rules for general-purpose AI models (the large models behind chatbots)
Aug 2, 2026 Most other rules, including transparency duties such as labeling deepfakes
Dec 2, 2027 / Aug 2, 2028 Rules for high-risk uses (such as hiring tools and credit scoring), delayed by 2026 amendments (Orrick)

In the United States there is no comprehensive federal AI law as of October 2026. A December 2025 executive order set up a Justice Department task force to challenge state AI laws (Paul Hastings), while states pass their own, such as Colorado’s law on automated decisions, effective 2027 (Skadden).

10.6 Jobs and the human in the loop

The evidence is mixed. From March through July 2026, AI was the most-cited reason for announced US job cuts in Challenger, Gray & Christmas’s monthly reports (about 24% of cuts announced through July), yet overall cuts fell and hiring plans rose. Challenger concluded that AI “is shifting the labor market, it is not dismantling it” (Challenger). Recent college graduates face a tough market (5.6% unemployment and 42% underemployment in mid-2026, per the New York Fed); how much of that AI causes is debated.

Human in the loop means a person reviews and approves AI work at key points and stays accountable for it. The skills that grow in value are the ones AI lacks: framing problems, knowing the business, judging quality, verifying, and taking responsibility. Learning to direct and check AI, the subject of this course, is itself a job skill.

Practice Questions

Easy

Question 1 · Multiple choice

According to the US Copyright Office, which of these can be protected by copyright?

  1. Code that an AI generated entirely on its own from a one-line prompt
  2. The human contributions to a work made with AI, such as choosing, arranging, and modifying the output
  3. Anything an AI produces, as long as you paid for the AI tool
  4. Nothing that involved AI at any stage

b. The Copyright Office’s position is that purely AI-generated material is not protected by copyright, while human contributions (choosing, arranging, and modifying the output) can be. Option a is purely AI-generated, so it is not protected. Option c is wrong: paying for a tool doesn’t change who created the material. Option d goes too far: using AI doesn’t remove protection for the parts a human contributed.

Question 2 · True or false

If a company’s website chatbot gives a customer wrong advice, the company is not responsible, because the chatbot made the mistake.

False. In 2024 a Canadian tribunal held Air Canada responsible for wrong refund advice from its website chatbot and rejected the idea that the chatbot was responsible for its own actions. A company is responsible for what its AI tells customers.

Question 3 · Short answer

What does human in the loop mean?

A person reviews and approves the AI’s work at key points (for example, approving an agent’s commands, reviewing code changes, checking a report before it is sent) and remains accountable for the result. The AI does much of the work; a human makes the important decisions and catches its mistakes.

Question 4 · Multiple choice

Which of these is the best general rule for using AI in your courses?

  1. If AI is allowed in one course, it is allowed in all of them
  2. Read the AI policy for each course and assignment, disclose AI use when required, and be able to explain everything you hand in
  3. AI use never needs to be disclosed, because everyone uses it
  4. Citations from an AI can be submitted as they are, because the AI found them

b. Each course sets its own rules, so you have to check the policy for every assignment, disclose AI use when it is required, and understand your own work. Option a is wrong because policies differ between courses. Option c ignores disclosure rules. Option d is dangerous: AI can invent citations, so never submit one you haven’t checked.

Medium

Question 5 · True or false

US courts have settled that training AI on copyrighted books without permission is legal.

False. The question is not settled. Anthropic agreed to a $1.5 billion settlement over books downloaded from pirate websites, and in 2026 an appeals court ruled in Thomson Reuters v. Ross that copying legal summaries to train a competing AI tool was not fair use, although that case involved a non-generative AI. Many other cases are still pending.

Question 6 · Short answer

What does it mean that the EU AI Act is risk-based? Give an example of a use at two different risk levels.

The rules depend on how much harm a use of AI could cause. Some uses are banned as “unacceptable risk” (for example, emotion recognition in the workplace). “High-risk” uses, such as AI for hiring or credit scoring, face strict requirements (risk assessment, data quality, logging, human oversight). Some uses only carry transparency duties (for example, labeling deepfakes), and most everyday uses (such as spam filters) have no specific obligations.

Question 7 · Scenario

Maria runs a small startup in the United States. A colleague says: “The EU AI Act is a European law, so it doesn’t concern us.” When could that be wrong?

The EU AI Act also applies to non-EU companies whose AI is used in the EU. If Maria’s startup offers an AI product or feature to customers in the EU, the Act’s rules for that kind of use (for example, transparency duties, or the stricter rules for high-risk uses such as hiring tools and credit scoring) can apply to it, even though the company is American.

Question 8 · Scenario

Sam uses a coding agent to build most of an app. Before he sells it, what should he check about ownership and licenses of the code?

  • Ownership: purely AI-generated material is not protected by copyright in the US, according to the Copyright Office; his own contributions (choosing, arranging, and modifying the output) can be.
  • Licenses: AI can produce code that closely resembles existing open-source code, which comes with license conditions such as keeping the original copyright notice. He should also check the licenses of every library the AI added.
  • Policy: if he builds it for an employer, he should follow the employer’s policy on AI-generated code.

Question 9 · Short answer

According to this page, which skills grow in value as AI does more of the work, and why?

The skills AI lacks: framing problems, knowing the business, judging quality, verifying, and taking responsibility. AI can produce a lot of output quickly, but a person still has to decide what problem to solve, check whether the result is right, and answer for it. Learning to direct and check AI is itself a job skill.

Hard

Question 10 · Long answer

A startup uses AI to write a feature. The code has a security hole that leaks customer data. Who is responsible: the AI vendor, the developer, or the company? Use an example from this page.

The company that shipped the code is responsible to its customers and regulators, and the developer (and whoever approved the code) is responsible within the company. The AI is a tool; “the AI wrote it” is not a defense. The Air Canada case shows the principle: a tribunal held the airline responsible for wrong advice from its own chatbot and rejected the idea that the chatbot was responsible for its own actions. Courts have likewise penalized lawyers for filing AI-invented citations. The safeguard is human review: code review, testing, and security checks before release.

Question 11 · Scenario

For a written report in another course, a student asks an AI to draft the whole report, pastes it in with its citations, and submits it. List the problems with this, and describe a responsible way to use AI for the same assignment.

Problems: it may break the course’s AI policy (and so academic integrity rules); the citations may be invented or may not support the claims; the facts may be wrong or outdated; the student may not be able to explain the work; and AI use was not disclosed.

Responsible approach: read the course policy first; if AI is allowed, use it for brainstorming, outlining, feedback, or explanations; write and understand the final work yourself; check every fact and open every source; and disclose how AI was used if required.

Question 12 · Long answer

“AI is destroying jobs.” Using evidence from this page, write a balanced response.

A balanced answer gives evidence on both sides:

  • Signs of disruption: from March through July 2026, AI was the most-cited reason for announced US job cuts (about 24% of cuts through July), and recent college graduates face high unemployment (5.6%) and underemployment (42%).
  • Signs against “destroying”: overall announced cuts fell and hiring plans rose, and Challenger concluded AI is “shifting the labor market, it is not dismantling it.” Economists debate how much of the graduates’ difficulty is caused by AI.

A good conclusion: AI is changing which tasks and skills are valued (judgment, problem framing, verification, domain knowledge, directing AI) more clearly than it is eliminating work overall, and the picture may change quickly.

Question 13 · Scenario

A startup wants to answer customer-support emails with the cheapest AI API it can find, run by a provider in another country. The emails contain names, addresses, and order details. What should the founders check before going ahead? Use ideas from this page and from Limits, Risks, and Responsible Use.

  • Data and privacy: where the data is sent and stored, whether the provider trains on it or keeps it, and whether privacy laws or customer contracts allow sending personal data to that provider and country.
  • Terms and security: the provider’s business terms, security practices, and reliability.
  • Quality and bias: test on realistic (anonymized) emails; check for wrong or inappropriate answers.
  • Prompt injection: customer emails are untrusted content, so the AI should not be able to take risky actions (refunds, account changes, sending data) on its own.
  • Human in the loop: a person should review replies, at least at first, because the company is responsible for what its AI tells customers (as Air Canada learned).
  • Regulation: if they have EU customers, check the EU AI Act’s transparency duties.

Challenge

Question 14 · Hands-on

Find your school’s (or this course’s) policy on using AI. Summarize three things it allows, three things it forbids or restricts, and what it says about disclosing AI use.

Answers depend on the school and course. A good answer quotes or links the actual policy, separates “allowed”, “restricted”, and “disclosure” rules clearly, and notes any differences between courses or assignments (many schools leave the details to each instructor).

© 2025-2026 Y. Rosenthal. All rights reserved.