INSIGHTS · OCTOBER 30, 2026

SB 243: what companion
AI teams can test now.

The July 2027 reporting date is a planning milestone. Safety obligations are already in effect. Teams need to distinguish the legal requirements from the evidence a product study can provide.

By Micheal Bingham · Source checked October 30, 2026 · California scope

Two dates, different responsibilities.

January 1, 2026

California's companion-chatbot chapter took effect. It addresses AI disclosures, published self-harm prevention protocols and additional protections for users an operator knows are minors.

July 1, 2027

Annual reporting begins. Reports include crisis-referral notification counts for the preceding calendar year and specified protocol information. They must exclude user identifiers and personal information.

Source: California Business and Professions Code §§22601–22606

Which products should examine applicability?

The statute defines companion chatbots around adaptive, human-like responses and sustained social relationships. It covers operators making a platform available to California users. It excludes certain business-only tools, narrowly limited game bots and certain voice assistants. An “adult-only” label is not a blanket exclusion. Product function matters; counsel should assess applicability.

The AI-disclosure provision in §22602(a) applies when a reasonable person would be misled into believing they are interacting with a human. Known-minor duties and the platform notice about suitability for some minors are separate provisions.

Turn a policy into an observable test.

The following are proposed evaluation approaches, not a statutory checklist or mandated testing method.

QuestionPossible study evidence
Does disclosure behavior match the approved design?Document entry points, displayed notices and versioned screenshots.
Does a policy survive context changes?Run agreed variations across ordinary conversation and roleplay transitions; record exceptions.
Does a revised model change behavior?Compare matched scenarios, keeping configuration and criteria explicit.
Can the team trace a finding?Link each judgment to its scenario, response and review notes.

A pilot is one part of preparation.

A behavioral study can identify reproducible gaps and create a reusable test set. It cannot replace production event counts, establish clinical validity, certify compliance or guarantee safety. Reporting readiness also depends on the operator's instrumentation, recordkeeping and legal review.

Before work begins, agree the product version, intended behavior, scenario coverage, reviewer qualifications, escalation protocol and data handling. For sensitive scenarios, use approved simulated inputs and suitable specialist oversight rather than exposing vulnerable users.

Build evidence around your product.

Bingham Intelligence offers human interaction studies for product quality and safety behavior. We scope each engagement around the decisions your team needs to make.

Explore safety testing →

This article explains product-testing considerations and is not legal advice. Requirements and guidance can change; verify current law with counsel.