Sanctions for fabricated citations, lawsuits over incorrect policy information, regulatory consequences for financial misinformation. Mata v. Avianca (2023), Air Canada (2024).

Wrong medical guidance on drug interactions, dosing, or diagnosis delivered with clinical-sounding confidence. Regulators are actively investigating medical AI chatbot claims.

Fabricated market data, hallucinated regulatory requirements, or invented research driving investment and business decisions.

Insecure code patterns recommended with confidence. Buffer overflows, SQL injection, path traversal — generated without warning. Pearce et al. (2022): 40% of Copilot’s security-sensitive outputs were vulnerable.

Hallucinated package names installed by developers, pre-loaded with attacker malware. Hallucination as a delivery mechanism (slopsquatting, 2024).

Public-facing chatbots giving wrong answers at scale. Every wrong answer is a potential media story, viral post, or tribunal ruling.