Not human, not optional: The new wave of state AI chatbot laws

What’s changing and why it matters

What’s new is not the concept of transparency; it’s the operational burden. States are drafting chatbot laws that reach into the product: how you disclose AI identity, how you handle high-risk conversations, what you log, and what you can prove after an incident. That’s a meaningful shift and it matters now because the first states to move are creating a copy and paste playbook, turning compliance from a single memo into a cross-functional project spanning product decisions, engineering work, vendor management, and executive risk calls.

Most proposals fall into three buckets, and each bucket drives a different set of build decisions:

1) Companion/emotionally responsive chatbots: Safety plus recurring disclosures

This bucket is about chatbots that feel personal. Legislatures are focusing on emotionally responsive “companion” features because they can keep users engaged, influence behavior, and reach minors. Recent Pacific Northwest proposals (including Oregon SB 1546 and Washington HB 2225) point in the same direction: Make it obvious it’s AI (and keep reminding users during long sessions), add guardrails for minors, and build a real self-harm escalation path (not just a disclaimer).

The business risk here is direct: These proposals can create obligations either you engineered into the product or you didn’t. And where private lawsuits or injunctive relief are on the table, gaps can turn into emergency fixes, unhappy customers, and reputational damage. 

2) High-impact decision chatbots: Governance when AI influences outcomes

This bucket is about when chat starts steering outcomes. If a chatbot meaningfully influences eligibility, ranking, access, or pricing in areas like employment, credit, housing, healthcare, or education, expect regulators to ask the hard questions: What is it supposed to do? How was it tested? How do you monitor drift? What happens when it’s wrong? Colorado’s SB 24 205 is one early model, but the larger trend is governance—documentation, monitoring, and escalation paths that are credible in the real world.

3) Employment use cases: Bias audits plus notices

Employment deserves its own callout because it’s where AI rules have already become operational. NYC Local Law 144 put bias audits and notices on the map for automated hiring tools, and similar expectations are spreading through customer diligence and regulator attention. If your chatbot is used for screening, interviewing, scoring, or recommendations, assume you’ll need to explain how it works, how it was assessed, and how people can challenge outcomes.

What this means for your product team

A recurring theme is proof. It won’t be enough to say you have a policy; you’ll need to show what you shipped, what you tested, what you monitored, and what you did when something went wrong. That means logging decisions, clarifying ownership, and keeping the kind of documentation that customers (and regulators) actually ask for.

Actionable takeaways

• Take a chatbot inventory that a regulator would recognize. List what you have, who uses it, what it’s for, and whether any features look “companion-like” or touch high-impact decisions.
• Set your disclosure standard now. Decide what the AI identity notice says, where it appears, and how you’ll handle repeat reminders during longer interactions.
• Engineer your “safety moments.” Decide how the bot responds to self-harm cues and minors, who owns escalation, and what you will (and won’t) log.
• Map where chat can change someone’s outcome. If chatbot output influences hiring, credit, housing, healthcare, or education decisions, treat governance as part of the product—testing, monitoring, and a clear path for review when the stakes are high.
• Design for change. Multistate rules will keep shifting; plan for feature flags, configurable disclosures, and a release process that can absorb new requirements without last-minute fire drills.

__________

If you have any questions about this Legal Briefing, please feel free to contact any of the attorneys listed or the Eversheds Sutherland attorney with whom you regularly work.