Move Fast and Mend Things: Keeping Up with Evolving AI Harms Using Social Media Commentary

arXiv:2610.09082v1 Announce Type: cross Abstract: The rapid deployment of AI systems has created socio-technical, psychological, and operational harms that can elude ex-ante threat modelling and ex-post incident tracking. We introduce an LLM-assisted thematic analysis pipeline to dynamically detect, categorise, and track emerging AI harms from large-scale social media data. Applying it to 5.7 million Reddit post summaries over 18 months (01/2025 to 06/2026), we curate and release a dataset of 575,000 AI harm-related posts and a bottom-up AI harm taxonomy of 12 categories and 47 subnodes. The taxonomy reliably covers established expert-defined risks while surfacing granular harms that top-down frameworks overlook, such as distinct forms of AI privacy violations. Temporal analysis surfaces evolving user-centric harms, such as agentic privacy and security breaches, premature AI adoption in the workplace, and grief from AI companion discontinuation. Our pipeline shortens harm-detection timelines and hereby complements efforts towards more participatory and responsive AI governance.

Discusses AI harms and their detection

This is a methodological contribution documenting real, varied AI harms reported by users at scale, which is useful for governance but also evidences meaningful negative societal impacts already occurring. The paper itself is neutral in stance but the underlying findings skew mildly negative.

The article reports on a method to track a wide range of AI harms dynamically, demonstrating that such harms are emerging and evolving. While the pipeline itself is a positive development for governance, the core implication is the existence and proliferation of these harms across society, including new forms like 'agentic privacy' and 'AI companion discontinuation grief.'

The paper identifies specific harms caused by AI systems, such as privacy violations and workplace issues. These harms are concerning for human flourishing. The pipeline's ability to detect and track these harms is a positive development for governance.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   f6e0467b-a414-43aa-9318-4b7ecbc8bb48
           https://arxiv.org/abs/2610.09082
   ↓
EVIDENCE   3 extracted excerpts
   ↓
MODEL RUN  4 runs, methodology 1.0.1
   ↓
SCORE      -79  (Strongly adverse)
   ↓
CONFIDENCE 74%