University of St. GallenFuture Society Hub
Automatically translatedDeutsches Original
Expert Discussion “Social Media goes Social”

Regulating
social media platforms.

Social media can be shaped through concrete techno-legal interventions.

Prof. Dr. Aurelia Tamò-Larrieux and Prof. Dr. Simon Mayer

University of St. Gallen
Konrad Adenauer Foundation, 4 November 2026

futuresociety.ch/event-slides/
kas-social-media-goes-social/index.en.html
Automatically translatedDeutsches Original
Who we are

Computer science meets law.

Simon Mayer

Prof. Dr. Simon Mayer

Professor of Computer Science

University of St. Gallen

Google Scholar ↗
Aurelia Tamò-Larrieux

Prof. Dr. Aurelia Tamò-Larrieux

Professor of Law & Technology

University of St. Gallen

Google Scholar ↗
Future Society Hub Logo
Future Society Hubfuturesociety.ch ↗
CoCoDa
snsf-cocoda.github.io ↗
Automatically translatedDeutsches Original
01 / Regulatory approaches

Where regulation can intervene.

The four regulatory levers, illustrated with a social media feed All content passes through the platform’s recommendation algorithm (lever III), which filters it step by step and optimises it for the KPI user engagement (lever IV). The filtered feed appears on a smartphone with infinite scroll (lever III). A person reaches for the smartphone; a clock stands for access barriers such as time or age limits (lever II). The person’s interactions flow back as data to the algorithm and, via sharing and reacting, to other users, whose new content enters the pool again. Systemic risks (lever I) arise for the individual, in the feedback loop and in the system as a whole, marked by a frame around the entire figure. All content Platform recommendation algorithm Profiling Ranking Selection KPI ≈ User Engagement optimises for For you Infinite scroll Access barriers Interaction data Share & reactCommunity new content Feedback loop III II IIIIII IV
ISystemic risksIIAccess barriersIIITargeted designIVPlatform KPIs
Automatically translatedDeutsches Original
02 / Lever I

Transparency creates evidence.
But platforms grade themselves.

ISystemic risks
IIAccess barriers
IIITargeted design
IVPlatform KPIs

Platforms assess and mitigate their own risks.

  • Annual risk assessment: discourse, elections, mental health
  • Data access for researchers (Art. 40)
  • Annual independent audit (Art. 37)
  • Very large platforms only
+
  • First causal feed experiments during an election campaign, thanks to platform data1
  • Simulated accounts independently measure what feeds serve2
−
  • Risk reports remain vague and lack evidence3
  • Research data access delivers incomplete data4
Enforce existing law consistently.
1Guess et al. 2023, Science2Bekavac et al. 2026, ACM TORS3ECNL 20264Bekavac & Mayer 2026, arXiv
Automatically translatedDeutsches Original
03 / Lever II

Age limits are visible.
Their effect is limited.

ISystemic risks
IIAccess barriers
IIITargeted design
IVPlatform KPIs

Keeps certain groups away from certain feeds without regulating the mechanisms.

  • Minimum age with age verification
  • Privacy-friendly: EUDI Wallet, zero-knowledge proofs
  • Time locks, e.g. at night: Meta settlement in the US1
  • Australia: ban for under-16s since December 2025
  • KIDS Act: no access under 13, parental accounts up to 152
+
  • Simple and politically visible
  • Access to Facebook worsened students’ mental health3
−
  • Australia: usage only drops from 78 to 69 %4
  • Age checks put data at risk: around 70,000 ID photos exposed5
  • Digital technology use explains at most 0.4 % of well-being6
Privacy-friendly. Not tied to age. A complementary measure
1Connecticut Attorney General 20262European Commission 2026, KIDS Act3Braghieri et al. 2022, American Economic Review4Barnes et al. 2026, BMJ5Discord 20256Orben & Przybylski 2019, Nature Human Behaviour
Automatically translatedDeutsches Original
04 / Lever III

Small design decisions.
Measurable consequences.

ISystemic risks
IIAccess barriers
IIITargeted design
IVPlatform KPIs

Regulates concrete mechanisms of the interface and the algorithm.

  • Interface: Endlos-Scroll, Autoplay, Likes, Push
  • Algorithm: ranking, feed without profiling, public value
  • Fixed requirements or open interfaces
  • Tested: Amsterdam1,2, USA3,4, China5; a reverse “Brussels effect”6
+
Interface
  • Batched notifications: better mood7
  • Without autoplay: greater sense of agency8
Algorithm
  • Chronological feed: less time spent9
  • Re-ranking via browser extension: less polarisation10
  • A feed without profiling is mandatory11, courts enforce the choice2
−
  • Evidence only for individual interventions
  • Interferes with economic freedom
  • Fixed requirements lag behind new features
Regulating does not mean prescribing code. What matters is the effect, and that can be tested12,13.
1Rechtbank Amsterdam 20252Gerechtshof Amsterdam 20263Connecticut Attorney General 20264New York State 20265Cyberspace Administration of China 20246Inverarity et al. 2026, ODI7Fitz et al. 2019, Computers in Human Behavior8Lukoff et al. 2021, CHI9Guess et al. 2023, Science10Piccardi et al. 2025, Science11Digital Services Act, Regulation (EU) 2022/206512Bekavac et al. 2026, ACM TORS13Guitton et al. 2022, Law, Innovation and Technology
Automatically translatedDeutsches Original
05 / Lever IV

Reverse the burden of proof.
Incentives for good design.

ISystemic risks
IIAccess barriers
IIITargeted design
IVPlatform KPIs

Targets the optimisation goals: today, engagement and advertising revenue.

  • New metrics: well-being instead of time on platform
  • Liability, duty of care, levies
  • Separate feed and hosting1
  • Public platform as an alternative2
Today

Society and researchers must prove that a design causes harm.

  • Effects are small and hard to attribute to a single feed3
  • Liability barely bites, because engagement drives both profit and harm2
With lever IV

Platforms show that their design is good, and are rewarded for it.

  • Align KPIs with societal goals, e.g. well-being instead of time on platform
  • Incentives instead of bans
There is evidence that the target metric matters: engagement ranking amplifies hostility4, and X’s algorithm shifts attitudes5.
1Keller 2021, Journal of Democracy2Mueller-Bloch & Ciriello 2026, CACM3Piccardi et al. 2025, Science4Milli et al. 2025, PNAS Nexus5Gauthier et al. 2026, Nature
Automatically translatedDeutsches Original
06 / Combination

The levers
work together.

Target zone
Strong evidenceWeak evidenceFast effectSlow effect
IIITargeted design
IVPlatform KPIs
ISystemic risks
IIAccess barriers
I
Systemic risks

Enforce now

Complete risk reports, real data access, external audits.

II
Access barriers

Complementary

Privacy-friendly, not tied to age.

III
Targeted design

Use selectively

Where evidence exists. Generate new evidence in Germany.

IV
Platform KPIs

Prepare for the long term

Assess moderate variants now.

A “right to customisation”1 connects the levers: users choose variants of a service whose effects become measurable.
1Tamò-Larrieux et al. 2021, Privacy Technologies and Policy
Automatically translatedDeutsches Original
07 / What do we do now?

What do we do now?
A proposal.

The levers help to bring concrete measures into existing law.

01 / Lever I

Feed inspection

Measure and audit feeds independently. Those who pass receive an inspection sticker.

02 / Lever III

Not addictive by default

Engagement lures off by default.

03 / Lever III

Visible trade-offs

Make informed choices between feeds.

04 / Lever IV

Advertising levy with a bonus

Safe design pays less.

Education for all Everyone understands how recommender systems work.
Automatically translatedDeutsches Original
08 / Proposal 1

Structured data collection
and feed inspection (“Feed TÜV”).

→Independent body with a mandate and a budget
→Simulated accounts measure what feeds serve, e.g. to a 14-year-old girl in Leipzig
→Feed report monthly, weekly before elections
→Inspection sticker with an expiry date, like a periodic vehicle inspection
Why measure independently?
27 of 51

data access requests rejected1

0 %

of problematic posts appear in the research APIs2

Only measures, deletes nothing, discloses methods and data.
1DSA40 Data Access Collaboratory 20262Bekavac & Mayer 2026, arXiv
Automatically translatedDeutsches Original
09 / Proposal 2

Not addictive
by default.

OFF BY DEFAULT
Push notificationsMessages from real people remain allowed, batched at fixed times on request4
Dark PatternsEngagement lures and manipulative design, e.g.:
→Lure notifications“You missed something!”
→Streaks with loss threats“Come back to the platform, or else …”
→Artificial countdowns“Only 9 seconds left to …”
→Confirmshaming“No thanks, I’d rather miss out.”
→Nagging“Are you sure you don’t want to …?”
95 %

of popular apps use dark patterns1

EU, 2026: TikTok’s design preliminarily found in breach of the DSA2

Meta, 2026: night-time locks for minors3

Off by default, changeable only through an active decision. Enforcement via feed inspection, collective actions and the Digital Fairness Act.
1Di Geronimo et al. 2020, CHI2European Commission 2026, TikTok3Connecticut Attorney General 20264Fitz et al. 2019, Computers in Human Behavior
Automatically translatedDeutsches Original
10 / Proposal 3

Visible trade-offs.
Informed choice.

Engagement feeds can amplify extreme content1. Anyone choosing between feeds should be able to choose in an informed way.

Feed without reshared posts2
+less misinformation
−less news
Chronological feed3
+fewer insults
−more unreliable sources
Less profiling4
+less data analysis
−less relevant ads
Bridging and public-value feeds5: selectable, but not mandatory.
Next step: a large-scale study on ranking methods.
1Milli et al. 2025, PNAS Nexus2Guess et al. 2023b, Science3Guess et al. 2023a, Science4Meta 20245Ovadya 2022
Automatically translatedDeutsches Original
11 / Proposal 4

Advertising levy
with a bonus.

→Flip the logic: Instead of proving that platforms cause harm, give them an incentive to do better
→Levy on the advertising revenue of very large platforms in Germany, as in Austria
→With a valid feed inspection sticker the rate drops significantly
→Revenue goes to education, research and feed inspection
Rate
High (e.g. 5–10 %)

cf. Austria and the proposed German “platform solidarity levy”

Low (e.g. 1–3 %)

with a valid inspection sticker

Safe design pays less.
Automatically translatedDeutsches Original
12 / Proposal 5

Education on recommender systems:
reach everyone.

01

For everyone

From primary school to adult education.

02

Compact

30 to 45 minutes, standardised, openly accessible materials.

03

Concrete

How a feed selects, what it optimises for, how engagement lures work.

Automatically translatedDeutsches Original
13 / Summary

Focus on the product,
not on bans.

ISystemic risks
IIAccess barriers
IIITargeted design
IVPlatform KPIs

Feed inspection

Measure and audit independently.

Complementary

Privacy-friendly, not tied to age.

Not addictive by default

Engagement lures off.

Visible trade-offs in the feed.

Advertising levy with a bonus

Safe design pays less.

Education on recommender systems for everyone connects the proposals: technical feasibility × legal enforceability.
Automatically translatedDeutsches Original
14 / Outlook

New technologies,
new challenges.

01

The next platform

AI chatbots increasingly shape “artificially social” interactions. Since August 2026, ChatGPT Search has been a very large online search engine under the DSA.1

02

Familiar problems

Systemic risks and designs that lure users into more interaction: levers I–IV can be transferred.

03

Evidence is lacking

Scientific evidence is needed even more urgently here, especially on long-term effects.

With social media, the evidence stayed with the platforms for too long. With AI chatbots, independent bodies should be able to measure their effects from the start.
1European Commission 2026b
Automatically translatedDeutsches Original
CoCoDa / Thank you

Thank you.

Following the event, the Konrad Adenauer Foundation will distribute a white paper that adds further detail to our brief overview.

snsf-cocoda.github.io ↗futuresociety.ch ↗
CoCoDa: Addressing systemic risks of dominant online platforms
University of St. GallenFuture Society HubCoCoDa
Automatically translatedDeutsches Original
Appendix

References.

Barnes, C. et al. (2026). Assessing early effects of Australia's Social Media Minimum Age Act on adolescents' social media use: observational study. BMJ 393, e363695.

Bekavac, L. & Mayer, S. (2026). Auditing Meta and TikTok Research API Data Access under Article 40(12) of the Digital Services Act. arXiv:2601.12390 (Preprint).

Bekavac, L., Strecker-Bischoff, J., Garcia, K., Mayer, S. & Tamò-Larrieux, A. (2026). Scrutinizing Systemic Risks in Personalized Recommender Systems Through Sock-Puppet Auditing of VLOPs. ACM Transactions on Recommender Systems 5(1), 1–47.

Boratto, L., De Conca, S., Fabbri, F. & Fabris, A. (2026). From Relevance to Reward: A Pipeline View of Addictive Design in Recommender Systems under the DSA. PMLR 350.

Braghieri, L., Levy, R. & Makarin, A. (2022). Social Media and Mental Health. American Economic Review 112(11), 3660–3693.

Connecticut Attorney General (2026). Attorney General Tong Announces $17.1 Billion Settlement with Meta, Forces Sweeping Child-Safety Reforms on Instagram and Facebook. Press release, 26 August 2026.

Cyberspace Administration of China (2024). Guidelines for the Construction of Minor Mode in the Mobile Internet (移动互联网未成年人模式建设指南), 15 November 2024.

Di Geronimo, L., Braz, L., Fregnan, E., Palomba, F. & Bacchelli, A. (2020). UI Dark Patterns and Where to Find Them: A Study on Mobile Applications and User Perception. CHI '20.

Discord (2025). Update on a Security Incident Involving Third-Party Customer Service.

DSA40 Data Access Collaboratory (2026). DSA40 Tracker Insights. As of 8 September 2026.

ECNL (2026). Five lessons from three years of risk assessments under the Digital Services Act. EDRi, 18 March 2026.

European Commission (2026a). Commission preliminarily finds TikTok's addictive design in breach of the Digital Services Act. Press release IP/26/312.

European Commission (2026b). Commission designates ChatGPT, Reddit, Roblox under Digital Services Act. Press release IP/26/1772.

European Commission (2026c). EU KIDS Act: helping children navigate a safer online world. 17 September 2026.

Fitz, N. et al. (2019). Batching smartphone notifications can improve well-being. Computers in Human Behavior 101, 84–94.

Gauthier, G., Hodler, R., Widmer, P. & Zhuravskaya, E. (2026). The political effects of X's feed algorithm. Nature 652, 416–423.

Gerechtshof Amsterdam (2026). Bits of Freedom v. Meta, appeal judgment, ECLI:NL:GHAMS:2026:594.

Guess, A. et al. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science 381, 398–404.

Guitton, C., Tamò-Larrieux, A. & Mayer, S. (2022). A typology of automatically processable regulation. Law, Innovation and Technology 14(2), 267–304.

Inverarity, C., et al. (2026). Bye-bye Brussels Effect. Hello, Beijing Effect! Open Data Institute, Blog, 29 September 2026.

Kaushal, R., van de Kerkhof, J., Goanta, C., Spanakis, G. & Iamnitchi, A. (2024). Automated Transparency: A Legal and Empirical Analysis of the Digital Services Act Transparency Database. FAccT '24, 1121–1132.

Keller, D. (2021). The Future of Platform Power: Making Middleware Work. Journal of Democracy 32(3), 168–172.

Kushlev, K., Proulx, J. & Dunn, E. W. (2016). "Silence Your Phones": Smartphone Notifications Increase Inattention and Hyperactivity Symptoms. CHI '16, 1011–1020.

Luguri, J. & Strahilevitz, L. J. (2021). Shining a Light on Dark Patterns. Journal of Legal Analysis 13(1), 43–109.

Lukoff, K. et al. (2021). How the Design of YouTube Influences User Sense of Agency. CHI '21.

Milli, S. et al. (2025). Engagement, user satisfaction, and the amplification of divisive content on social media. PNAS Nexus 4(3), pgaf062.

Mueller-Bloch, C. & Ciriello, R. (2026). Governing Social Media as a Public Utility. Communications of the ACM 69(10).

New York State (2026). Governor Hochul and Attorney General James Announce Final SAFE for Kids Act Rules to Protect Children Online.

Orben, A. & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour 3, 173–182.

Ovadya, A. (2022). Bridging-Based Ranking. Belfer Center for Science and International Affairs, Harvard Kennedy School.

Ovadya, A. & Thorburn, L. (2023). Bridging Systems: Open Problems for Countering Destructive Divisiveness across Ranking, Recommenders, and Governance. arXiv:2301.09976 (Preprint).

Piccardi, T. et al. (2025). Reranking partisan animosity in algorithmic social media feeds alters affective polarization. Science 390, eadu5584.

PRO Medienmagazin (2026). Bundesregierung plant schärferen Jugendschutz in sozialen Medien [Federal government plans stricter youth protection on social media].

Rechtbank Amsterdam (2025). Bits of Freedom v. Meta, judgment in summary proceedings (kort geding) of 2 October 2025, C/13/774725 / KG ZA 25-687.

Tamò-Larrieux, A., Zihlmann, Z., Garcia, K. & Mayer, S. (2021). The Right to Customization: Conceptualizing the Right to Repair for Informational Privacy. Privacy Technologies and Policy, Springer.

Wagner, M. W. (2023). Independence by permission. Science 381, 388–391.

All slides · G or Esc to close