Format. Break into groups of 4–5. Each group picks one of the two breakouts below (or takes both if time allows). Spend ~5 minutes skimming the prep reads, then ~10 minutes debating. A designated reporter brings the group’s position — and any dissents — back to the full class for a ~3-minute report-back.
Section 3.3.7 is careful to distinguish misinformation (false, not necessarily malicious) from disinformation (deliberately deceptive) — but flags that intent is nearly impossible to determine from content alone. The book also argues that LLMs and deepfakes change the economics of disinformation, not its structure. Both breakouts push on what platforms and regulators can actually do about that.
Motion. “Meta’s January 2025 decision to end its US third-party fact-checking program and replace it with X-style community notes was a legitimate governance choice, not a capitulation.”
Prep reads (5–10 min).
Discussion prompts.
- The book’s central worry is that disinformation is most damaging during periods of uncertainty (elections, pandemics), when speed to correction matters most. Community Notes require multi-perspective consensus before display — which can take hours or days. Is that a bug or a feature?
- Third-party fact-checkers were routinely accused of political bias (from both left and right). Community Notes are crowdsourced. Which is more likely to produce durable legitimacy — expert authority or bipartisan consensus? Which produces more accuracy?
- The Argentina 2023 deepfake in the book was viewed millions of times before being debunked. Would Community Notes have caught it faster? Slower? Not at all?
- If Community Notes work well on X (an open question), does that mean fact-checking was always a fig leaf — or does it mean X’s user base happens to include enough contrarians to make consensus meaningful, and the model won’t transfer?
Bring back. Your group’s verdict — “legitimate replacement,” “fig leaf,” or “depends on X” — and the single metric you’d use to judge which.
Breakout B: Liability for Algorithmic Amplification
Motion. “Platforms should face legal liability when their recommendation algorithms amplify demonstrably false content about elections or public health, even if they did not create the content themselves.”
Prep reads (5–10 min).
- Commission fines X EUR 120 million under the Digital Services Act — European Commission, December 2025. First DSA non-compliance fine; hits X on deceptive design (blue checks), ad-repository transparency, and researcher access — a template for amplification liability.
- EU’s Preliminary DSA Findings Put TikTok’s Engagement Design in the Regulatory Crosshairs — ComplexDiscovery, February 2026. Analyzes the Commission’s preliminary finding that TikTok’s infinite scroll, autoplay, and personalized recommender create Article 34 systemic risks.
- Section 230 at 30: Internet’s Legal Shield Faces Coordinated Repeal Push and Trial Threats — The Meridiem, February 2026. Surveys the Durbin-Graham sunset bill, Third Circuit’s TikTok “blackout challenge” ruling, and two March 2026 rulings holding recommender output outside Section 230.
- Systematic partisan content skews in TikTok during the 2024 US elections — Nature, May 2026. NYU-Abu Dhabi audit of 394k videos across 323 sock-puppet accounts: For-You feed favored Republican-aligned content ~12% over Democrat-aligned regardless of seeding — a concrete amplification-liability test case.
Discussion prompts.
- The book argues LLMs make it “cheap to produce many high-quality variants of a message.” If liability attaches to amplification, does it force platforms to downrank AI-generated content wholesale? What does that do to legitimate uses?
- Section 230 currently treats platforms as distributors, not publishers. But the recommendation algorithm is arguably a platform’s own speech. Where should the line sit between hosting user content (immune) and recommending it (liable)?
- “Demonstrably false” sounds clean, but the book flags that intent is rarely observable and content is often mixed (a true claim in a misleading frame). What’s a workable legal standard — and who adjudicates?
- The EU’s DSA already imposes systemic-risk assessments on very large platforms. Is that the right model, or does it just entrench incumbents (per the book’s “entrenchment effect” argument)?
Bring back. The narrowest liability rule your group could support, and the strongest First Amendment / free-expression objection to it.
Instructor notes
These breakouts map to Section 3.3.7’s takeaways that (1) intent is rarely observable, so moderation must focus on harm and scale, and (2) short-lived falsehoods still corrode trust. Breakout A is the more current — Meta’s January 2025 pivot is fresh and students often have strong priors on both sides. Breakout B is where the technical and legal literacy pay off; students who read the DSA sections of Chapter 3 will engage more deeply. If time is very short, run A only; it produces sharper disagreement.