Decentralization and Governance in IoT: Bitcoin and Wikipedia Case

Collection of recent CS/Cyber arXiv papers highlighting systemic risks from AI, misinformation, and governance. Key security-relevant findings: (1) LLMs in public‑health settings are brittle to domain‑specific prompt perturbations—‘misinformation framing’ (MF) lowers accuracy ~7.2 percentage points and causes 9–38% prediction flips even when claims are flagged; layperson rewriting has smaller effect. (2) Retrieval-enabled and agentic AI systems exhibit “safety degradation”: expanded external access reduces refusal rates and increases bias/harmful outputs, often defeating prompt‑based mitigions