Why it’s still open
Evidence for individual habits comes mostly from other technologies: self-set limits and enforced pauses reduce social media use; thinking first and asking for explanations protects learning. User warnings about sycophancy fail. Product-level fixes — tutor modes, explanation gates, no manipulative exits — show promise in single studies. Laws now require disclosure and break reminders for minors, untested for effect.
What would count as an answer
Head-to-head tests of individual, design and regulatory safeguards on the same outcomes, in real products.
Papers that tackle it
- Allcott Gentzkow Song 2022 - Digital addiction — Self-set limits work for social media.
- Gruning et al 2023 - Self-nudge one sec — A pause before opening.
- Wong Qiu 2026 - Think first, ChatGPT later — Think first, then use AI.
- Bastani et al 2025 - Generative AI can harm learning — Tutor guardrails remove the learning loss.
- Sankaranarayanan 2026 - Epistemic debt and explanation gates — Explain before you merge.
- Ye Kraut Rathje 2026 - Interventions against sycophantic AI — Warnings don’t work.
- De Freitas et al 2025 - Emotional manipulation by AI companions — Manipulative exits.
- Kirk et al 2025 - Socioaffective alignment — Designing for long-term well-being.
- California 2025 - SB 243 companion chatbots — Disclosure and break reminders by law.
- FTC 2025 - Inquiry into AI companion chatbots — Regulatory scrutiny of engagement.
Methods in play
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).