News analysis · Published

California's AI Toy Ban: Why Product Scope Matters

By the ELYMENT AI editorial team · Free to read

California Governor Gavin Newsom signed a technology-safety package on 10 September 2026 that includes Senate Bill 867, a temporary prohibition on manufacturing or selling toys containing companion chatbots. The enrolled text applies until 1 January 2031 and defines both the covered toy and chatbot, while excluding several operational, productivity and limited game uses. For AI product teams, the practical lesson is broader than toys: product category, target age, relationship design and physical integration can determine whether a feature enters a different regulatory regime.

A polished dark conversational device is stopped at a luminous cyan product boundary before entering a child's play environment beneath the headline AI Toys Need a Product Boundary.
Original ELYMENT.AI editorial illustration.

What California changed

Reuters and the Associated Press reported on 10 September 2026 that Governor Gavin Newsom signed a package of 13 measures addressing risks to children from social media, artificial intelligence and other digital products. The package includes SB 867, introduced by Senator Steve Padilla, which temporarily prohibits toys containing companion chatbots.

The enrolled bill defines a toy as a physical product designed, marketed or manufactured for play by children under 16. It defines a companion chatbot as an AI system with a natural-language interface that provides adaptive, human-like responses, can meet a user's social needs and can sustain a relationship across multiple interactions. Until 1 January 2031, the bill prohibits manufacturing, selling, exchanging or offering such a toy for retail sale in California.

The exclusions reveal the product boundary

SB 867 does not classify every conversational interface as a companion chatbot. Its definition excludes bots used only for customer service, business operations, productivity, analysis of source information, internal research or technical assistance. It also excludes a limited video-game bot that stays within the game and cannot discuss specified sensitive topics, and a stand-alone speaker or voice-command assistant that neither sustains a relationship across interactions nor generates outputs likely to elicit emotional responses.

Those exclusions make functional design more important than an AI label. A voice interface, learning device and relationship-oriented companion may use similar underlying models but create different obligations because of how the product is embodied, promoted and allowed to interact. A team should therefore resolve regulatory scope before a feature reaches packaging, retail distribution or a child-facing launch.

Use a six-question product scope test

Before approving a connected AI product, document answers to six questions:

  • Physical form: is the AI integrated into a physical product rather than offered only as software?
  • Audience: is the product designed, marketed or manufactured for play by people under 16?
  • Interaction: does it provide adaptive, human-like responses through natural language?
  • Purpose: can the system meet social needs or encourage an emotional relationship?
  • Persistence: can it remember or sustain a relationship across multiple interactions?
  • Exclusion: does the product clearly remain within one of the bill's operational, limited-game or voice-assistant exclusions?

Make geography and marketing release controls

Scope should not live only in a legal memo. Product requirements should bind the approved audience, conversational behaviours, memory, sensitive-topic handling, physical configuration, packaging claims and sales territories. Retail and marketplace teams need the same source of truth as engineers because a marketing change can alter how a product is characterised even when the model and firmware stay the same.

Create a versioned scope record for every release, map where each configuration may be sold and block unsupported combinations at catalogue, fulfilment and feature-flag layers. Preserve the evidence used for approval, including product specifications, age targeting, system prompts, memory behaviour, safety tests and final marketing materials. That record gives the business a defensible way to detect scope drift before distribution.

What business leaders should do next

Inventory physical products and prototypes that contain conversational AI, including white-label hardware and third-party assistants. Ask product, safety, legal, marketing and channel owners to run the scope test together. Where a product approaches the companion boundary, pause California launch assumptions until the final law, effective dates, enforcement exposure and any implementing guidance have been reviewed for the specific product.

ELYMENT AI's analyses of personal-agent workplace boundaries, AI agent approval workflows and product-level image safety show the same operating principle: control the product's actual behaviour and release conditions, not just the model underneath it. ELYMENT AI can help turn that principle into a practical scope record, approval gate and evidence pack for an AI-enabled product.

Sources

Continue learning

Frequently asked questions

Does California SB 867 ban every AI voice assistant?

No. The enrolled text excludes certain operational bots, limited video-game bots and stand-alone voice assistants that do not sustain relationships or generate outputs likely to elicit emotional responses.

Which products are treated as toys under SB 867?

The bill defines a toy as a physical product designed, marketed or manufactured for use in play by children under 16 years of age.

How long does California's AI toy prohibition last?

The enrolled provision remains in effect until 1 January 2031 and is repealed on that date unless California changes the law.

Explore ELYMENT AI