← Social media bots, 2016–2026

How the governance frame shifted

Who gets named responsible, which harms surface which year, and which policy frame wins — read across a hand-curated 111-paper governance/legal set plus the full 3,233-paper corpus's governance tags.

111-paper governance set · 3,233-paper corpus·← corpus overview·literature story →

Who should govern bots?

Governance actor named, by yearfull corpus, stacked
In the 111-paper set, international/multilateral framing (EU, DSA, UN) now leads (34 papers) — ahead of national government (28) and platform self-regulation (25). Local/state government: 1 paper.

Harms & benefits, by year

Harm category mentions, by year111-paper governance set

Years active = distinct years at least one paper raised that harm.

Bots in the military, elderly, crisis response…

Subgroup / context focus111-paper governance set
Elderly users and public-health contexts are least-discussed (1–2 papers) despite being flagged as high-vulnerability elsewhere in the HCI/chatbot literature — a gap between where harm is studied and where it's plausible but under-researched.

The four policy-debate frames

Innovation · Security · Economy · Rightsby year

The "don't over-regulate" frame — the one most invoked in public industry debate — is the least-represented argument in the academic literature itself.

Harm type: physical, economic, informational, psychological, political, social

Papers by harm type111-paper set, totals

What a more nuanced governance would do

The two biggest governance-idea categories every year are blunt by design — Detection & Technical Countermeasures (48 papers, 2025) and General Countermeasures (54, 2025). What's growing faster, off a smaller base: Multi-Stakeholder Collaboration (0→7), Accountability & Liability (1→7), Ethical & Normative Design (1→9).

1. Disclosure, not a binaryTie detection to disclosed-vs-undisclosed, so conversational AI and influence bots stop being regulated identically.
2. Formalize the fastest-growing ideaMulti-stakeholder / co-governance categories are growing fastest from the smallest base — give them standing, not an afterthought.
3. Extend "international" downwardThe framing that already dominates platform governance should reach the under-studied subgroups — elder fraud, health-crisis misinformation.