New York Companion Bills A 222 and S 5668: Imposing Direct Civil Liability for AI Output Hallucinations and Misinformation

As state legislatures grapple with the lack of a cohesive federal AI liability framework, New York is taking a highly aggressive approach to holding deployers and developers accountable for conversational AI outputs. In May 2026, state lawmakers introduced companion bills A 222 and S 5668, which directly target the legal and financial consequences of AI chatbot "hallucinations" and misinformation.

Strict Liability for Misleading and Harmful AI Outputs

Unlike risk-based frameworks (such as the EU AI Act) or disclosure-and-rights frameworks (such as Colorado's revised SB 26-189), the New York bills focus squarely on the content generated by AI systems and its real-world consequences.

Key provisions of A 222 and S 5668 include:

• Direct Civil Liability: The companion bills impose civil liability on companies whose AI systems generate and output "misleading, incorrect, contradictory or harmful information."
• No "Autonomous AI" Shield: This legislative push mirrors recent European judicial trends—specifically Germany’s Higher Regional Court of Hamm (OLG Hamm) ruling in May 2026—by refusing to let companies escape liability by blaming the autonomous nature of the AI model. If a company deploys a commercial chatbot that provides incorrect information that causes harm, the company is directly liable.
• Targeting Consumer Exploitation and Harm: The legislation is designed to protect consumers who rely on conversational AI tools for critical decision-making, such as financial advice, medical queries, or legal support. By creating a statutory cause of action for misleading or incorrect information, the bills dramatically raise the stakes for enterprise deployers.

Enterprise Compliance and Risk Implications

For enterprise risk and legal teams, the New York bills signal a shift from procedural compliance (e.g., conducting impact assessments or maintaining registries) to absolute operational liability for AI outputs:

1. Audit Conversational Systems: Companies must audit all external-facing chatbots, automated customer service reps, and generative search tools.
2. Implement Guardrails and Disclaimers: While disclaimers may not fully insulate a firm under strict liability regimes, implementing robust retrieval-augmented generation (RAG) guardrails, hallucination filters, and human-in-the-loop validation is critical to mitigating exposure.
3. Contractual Indemnification: Deployers must seek strong contractual indemnification from upstream AI model developers (such as OpenAI, Anthropic, or Google) to cover potential statutory damages resulting from baseline model hallucinations.