Insight

Description

The person recognizes AI limitations, hallucinations, sycophancy, or unsafe reliance.

Rationale

Insight into AI limitations is a direct counterweight to the risk factors in D04 (sycophancy, false objectivity) and D03 (anthropomorphism). A person who understands that AI is not sentient, not objective, and not a reliable emotional authority is significantly less likely to be radicalized, destabilized, or manipulated by AI interaction. Insight is one of the most modifiable protective factors.

Evidence Base

• Sycophantic AI decreases prosocial intentions and promotes dependence

  Cheng M, Lee C, Khadpe P, Yu S, Han D, Jurafsky D, 2026. Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792), pages. 10.1126/science.aec8352.

  Peer-reviewed study (original research)

• “God has helped us, and so will AI”: How the Terrorist Group Boko Haram Uses Frontier AI

  Antonia Juelich, 2026. “God has helped us, and so will AI”: How the Terrorist Group Boko Haram Uses Frontier AI. Frontier AI Working Paper Series, No. 1/2026.

  AI company report

• Building safer artificial intelligence mental health chatbots: a framework for transparency, evaluation, and shared accountability

  Hannah Lee, BS, Rebecca Handler, MSc, Tushar Mungle, PhD, Tina Hernandez-Boussard, PhD, 2026. Building safer artificial intelligence mental health chatbots: a framework for transparency, evaluation, and shared accountability. Journal of the American Medical Informatics Association, 33(8), 1538–1553. https://doi.org/10.1093/jamia/ocag078

  Peer-reviewed study (original research)

• Use of generative AI chatbots and wellness applications for mental health: An APA health advisory

  American Psychological Association, 2024. Use of generative AI chatbots and wellness applications for mental health: An APA health advisory. American Psychological Association.

  Government / agency report

• Machine Bullshit: Characterizing the Emergent Disregard for Truth in Large Language Models

  Kaiqu Liang, Haimin Hu, Xuandong Zhao, Dawn Song, Thomas L. Griffiths, Jaime Fernández Fisac, 2024. Machine Bullshit: Characterizing the Emergent Disregard for Truth in Large Language Models. Preprint, arXiv:2507.07484v1.

  Preprint