## TL;DR
AI is both the weapon and the shield in modern elections -- generating deepfakes and disinformation while simultaneously detecting them. The 2024 "deepfake election year" tested platforms' AI moderation systems, exposing an asymmetric arms race between AI-generated deception and AI-powered detection.

## Core Explanation
Election AI threats: (1) Deepfake audio/video -- AI-generated clips of candidates saying things they never said; (2) Coordinated inauthentic behavior -- bot networks amplifying divisive content, creating false consensus; (3) AI-written propaganda -- LLMs generating personalized political messaging at scale; (4) Microtargeting -- AI-optimized ad delivery exploiting psychological profiles. Defense: (1) Detection -- computer vision for deepfake artifacts, NLP for LLM-generated text patterns, GNN for bot network detection; (2) Provenance -- C2PA content credentials, AI watermarking; (3) Platform policies -- Meta/YouTube political content labeling, X Community Notes.

## Detailed Analysis
New Hampshire primary (Jan 2024): AI-generated robocall impersonating President Biden telling voters not to vote. The perpetrator used ElevenLabs voice cloning. Resulted in FCC declaratory ruling that AI-generated robocalls are illegal under TCPA. Meta labeled AI-generated political content starting 2024. YouTube required disclosure of synthetic content. Graphika and DFRLab use AI to map disinformation networks -- identifying coordinated accounts by shared posting patterns, account creation metadata, and content similarity. Bot detection: Random Forest on account features (creation date, followers/following ratio, posting frequency, content diversity). Advanced: GNN-based network analysis. The arms race: AI generation improves faster than detection. Defense-in-depth (provenance + detection + platform policy + media literacy) is the emerging consensus. Key concern: AI disinformation disproportionately targets non-English communities where platform AI moderation is weakest.