security-course

Security, Privacy, and Consumer Protection

View the Project on GitHub noise-lab/security-course

Learning Objectives

By completing this assignment, you will:

Introduction

Copyright law governs much of what we can and cannot do online, from sharing memes to uploading videos to creating AI-generated art. But how do platforms like YouTube, TikTok, and Instagram actually enforce copyright? What survives and what gets taken down? Where are the boundaries of fair use in practice? And how, mechanically, does a platform recognize a copyrighted song or clip the instant you upload it?

In this assignment, you’ll conduct hands-on experiments with copyrighted content on a platform of your choice, investigate how AI-generated content is treated, explain the detection mechanism that decides the outcome of each experiment, and analyze the results through the lens of copyright law.

Important: This assignment involves deliberately testing platform copyright enforcement. Use good judgment:

Grading & Rubric (100 points)

This rubric is shown up front so you know where to invest your effort. Labs are graded primarily for thoughtful completion; points reward understanding, not polish.

Component Points What earns full marks
Platform policy analysis 14 You document how your chosen platform detects, flags, and adjudicates copyright (automated matching, reporting, appeals/counter-notification, monetization, licensing programs) and note what you’ll compare it against.
Fair-use experiments documented (table + outcomes) 18 A per-upload text table records content uploaded, time-to-detection, outcome, and options presented for 2–3 experiments spanning the transformativeness spectrum. Screenshots corroborate but every claim is in text.
AI-generated content investigation 14 For 2–3 AI pieces you record prompt, output, platform response, and ownership findings verified against the actual ToS (not an LLM summary).
Legal analysis (4 factors + case law + gap) 24 You apply all four fair-use factors to each experiment, cite relevant case law, and analyze the gap between law, policy, and enforcement.
Detection mechanism (depth) 15 You explain mechanistically how fingerprinting/content-matching works and use it to explain why your own experiments evaded or triggered detection.
Reflection & AI-verification 10 You report what you tried (including dead ends), what surprised you in your own experiments, and — if you used an LLM — at least one ownership/ToS claim you checked against the source and what you found.
Evidence completeness (screenshots/links/timestamps) 5 Appendix contains screenshots, links to uploaded content, and timestamps that corroborate the text.
Extra credit: cross-platform or detection-threshold experiment +10 Run the same content across 2–3 platforms, OR vary one transformation to find the threshold where matching breaks — reported as a small table. See the stretch task below.

Tie every gradable claim to your own uploads and outcomes. Generic prose that could describe anyone’s run earns little credit; the analysis must be grounded in your specific experiments.

Tasks

Choose ONE platform to focus on for this assignment. Options include:

Research and document the platform’s copyright policy:

Compare the platform’s stated policy with the behavior you observe in your experiments (Tasks 2 and 3).

2. Fair Use Experiments

Upload 2-3 pieces of content that test different aspects of fair use. Each piece should represent a different point on the spectrum of transformativeness. Suggested experiments:

For each upload, document — as a text table (screenshots corroborate, but the grader reads the table):

What I uploaded Time to detection Outcome Options presented
e.g., 8s unmodified pop-song clip immediate / minutes / hours / never up / muted / blocked in regions / removed / monetization diverted dispute / acknowledge / trim / replace audio

For each row also keep: screenshot of successful upload; screenshot of any warnings, flags, copyright claims, or takedown notices; and a note of the final outcome (stays up, muted, region-blocked, removed, monetization disabled, etc.). Paste the table into your report — every gradable claim must be in the text, not only in an image.

3. AI-Generated Content Investigation

Create 2-3 pieces of content using AI tools (e.g., DALL-E, Midjourney, Stable Diffusion, ChatGPT, Suno, etc.) with varying degrees of similarity to copyrighted works:

Upload each piece to your chosen platform and document:

Apply what you’ve learned about copyright law to your experiments:

5. Detection Mechanism (depth)

This is the missing technical layer. Explain how your platform actually detects copyrighted content, then connect that mechanism to what you observed in Tasks 2 and 3.

6. Reflection & Tinkering (required)

This is where you show the work is yours. In a short reflection (a few paragraphs):

7. Stretch — cross-platform or detection-threshold experiment (extra credit, +10)

Anyone can describe fingerprinting; prove it by probing the matcher empirically. Pick one:

Using AI (encouraged, with verification). You may use an LLM to help interpret a platform policy or draft a fair-use analysis. If you do, include the exchange in the appendix. For the AI-generated-content task especially, verify any claim about who owns AI output against the actual ToS (and U.S. Copyright Office guidance) rather than trusting the model’s summary — models routinely overstate or invent ownership terms. Quote the governing clause that confirms or contradicts the model. Submitting an assertion you can’t back up against the source will lose points; catching the model in an error will earn full marks for that item.

Be ready to defend it. Per the syllabus, we may ask you to reproduce or explain any part of this lab live (office hours, a pop quiz, or the exam) — e.g., “re-upload this clip and show the Content ID claim,” or “walk me through why your pitch-shifted version evaded the matcher.” Do the work so you can.

Optional Extensions

If you want to explore further (beyond the graded stretch above), here are some additional ideas:

Submission Instructions

Submit a single markdown report named copyright-report.md plus a folder of screenshots. Because your report is graded from its text, document every experiment in the text tables and prose described below — screenshots are corroboration, not a substitute for the text. Push the report and the screenshots folder to your private GitHub repository (do not push a zip file).

Your report must contain these headings, in this order (they map one-to-one to the rubric above):

# Copyright Lab — <your name>

## 1. Platform Policy Analysis
   (chosen platform; detection method, flagging, appeals/counter-notification,
    monetization, licensing programs; what you'll compare against)

## 2. Fair Use Experiments
   - Per-upload TABLE: content | time-to-detection | outcome | options presented
   - One row per experiment (2–3 experiments across the transformativeness spectrum)

## 3. AI-Generated Content Investigation
   - Per piece (2–3): exact prompt | output (screenshot) | platform response
   - Ownership findings VERIFIED against the actual ToS (quote the clause)

## 4. Legal Analysis
   - Four fair-use factors applied to EACH experiment
   - Relevant case law
   - Gap analysis: law vs. policy vs. enforcement (note: "not taken down" ≠ "legal")

## 5. Detection Mechanism (depth)
   - How fingerprinting/content-matching works; robustness; where it breaks
   - Mechanistic explanation of why YOUR experiments evaded or triggered detection

## 6. Reflection & Tinkering
   - What you tried that didn't work; what surprised you in YOUR experiments;
     one AI ownership/ToS claim you verified against the source

## 7. (Extra credit) Cross-platform or detection-threshold experiment
   - Small table; the threshold or comparison you found; tie back to the mechanism

## Appendix: screenshots, links, timestamps, and AI usage (if any)
   - All screenshots; links to uploaded content; upload/detection timestamps
   - Any AI prompts, model output, and your verification against the source

Resources

Academic Integrity Note

This assignment involves creating and uploading content that may be flagged or removed. This is done for educational purposes to understand copyright enforcement. Do not:

When in doubt, ask the instructor.