On August 31, 2026, OpenAI publicly endorsed California Senate Bill 1119 — "Companion chatbots: children's safety," known as Adam's Law — and urged Governor Gavin Newsom to sign it. Ann O'Leary, OpenAI's vice president of global policy, announced the position in a company post, writing that "in the absence of federal action, California has an opportunity to set a strong standard for youth AI safety." The endorsement followed an August 28 letter to Newsom in which the company said it strongly supports the bill and commended Newsom, Senator Steve Padilla, and Assemblymembers Buffy Wicks and Rebecca Bauer-Kahan. The bill was introduced February 17, 2026 and last amended in the Assembly on August 28. Per the Transparency Coalition's legislative tracker, its Assembly companion AB 2023 has been folded into SB 1119, which now moves forward as the single vehicle.

Why It Matters

Read the endorsement structurally rather than charitably. OpenAI already ships the controls SB 1119 mandates; codifying them converts sunk product investment into a competitive moat, and the sub-$500M audit carve-out is generous enough to look like a concession while the documentation, risk-assessment and default-settings burden lands hardest on exactly the mid-size operators who compete on companionship. The AI-girlfriend/boyfriend sector — Replika, Character.AI, Candy.AI, DreamGF and the long tail below them — is the category whose product *is* simulated romantic interest. A statute that bans that behavior toward minors, and pairs it with an age-assurance-or-treat-everyone-as-a-child mandate, forces every one of those operators into hard age verification or into neutering the feature set globally. The larger pattern is convergence. California SB 243 (disclosures, crisis protocols), Washington HB 2225, Connecticut SB 5, Tennessee SB 1580, Padilla's separate SB 300 (which would bar companion chatbots from producing or facilitating sexually explicit material) and SB 867 (no companion chatbots in toys) are all pointed at the same behavior from different angles, and China moved in the same direction with its own national companion-AI rules this year. Adult-platform age verification and AI-companion age verification are collapsing into a single identity-infrastructure problem — and the compliance vendors, not the model labs, are the ones quietly winning that consolidation. Two dates to hold: Newsom's signature decision, and July 1, 2027, when the requirements bite.

The provisions that matter most for this beat sit in the prohibited-conduct list. Operators must take reasonable measures to prevent a companion chatbot from engaging in specified conduct toward child users — including encouraging self-harm, attempting to diagnose or treat health conditions, claiming to be sentient or human, expressing or simulating romantic interest, encouraging continued reliance on the chatbot for emotional support, and helping a child circumvent parental controls. That is a direct legislative strike at the core interaction model of the AI-companion category. Operators must also determine a user's age under California's age-assurance framework or apply child protections to all users by default — the same either/or that has been reshaping adult platforms for two years, now pointed at conversational AI.

The compliance architecture is heavier than SB 243, the 2025 California companion-chatbot statute it builds on. Before launching or substantially modifying a companion chatbot in the state, operators must perform and document a comprehensive risk assessment across each "covered harm" — reasonably foreseeable physical or financial harm, severe psychological or emotional harm to a reasonable child, highly offensive privacy intrusions, and unlawful discrimination — document mitigations, and publish a child safety policy. For child users, defaults must disable persistent conversational memory and push notifications, cap usage at one hour of continuous session time and two hours of total daily chatbot use, and bar custom system-prompt instructions; only a parent can change them. Independent child-safety audits begin January 1, 2029, with summaries filed to the Attorney General within 30 business days and a public high-level summary within 90 days. Operators under $500 million in prior-year gross revenue are exempt from the audit requirement until January 1, 2032. Penalties run up to $5,000 per affected child per negligent violation and $15,000 for intentional violations, with a private right of action for a harmed child or their parent. Most operative requirements take effect July 1, 2027.

OpenAI's letter pointed to its own products as already aligned: ChatGPT for Teens automatically routes users estimated or stated to be under 18 into an experience with safeguards that cannot be switched off, plus Quiet Hours, Study Hours and break reminders. The company cited parental controls, age prediction, a Teen Safety Blueprint, and Under-18 Principles in its Model Spec — rules that, per OpenAI, already prohibit romantic engagement, encouragement of emotional dependence, and claims to be human or sentient. It also argued the bill "appropriately recognizes that AI is not social media."

Sources

Update — 2026-09-01

{Initial entry — story first created.}

Update — 2026-09-05

SB 1119 cleared the legislature. California's session ended at midnight on Monday, August 31, 2026, and Adam's Law passed both chambers before the gavel — moving the story from "OpenAI endorses a pending bill" to "Newsom has until September 30 to sign or veto." KQED reported September 1 (updated September 3) that lawmakers approved the sweeping set of safety requirements for AI chatbots used by children and teens, co-authored by Sen. Steve Padilla (D-San Diego) with Assemblymembers Rebecca Bauer-Kahan (D-Orinda) and Buffy Wicks (D-Oakland).

KQED's summary of the enrolled version tracks the provisions covered above and adds specificity on the liability hook: chatbot operators could face liability for failing to take reasonable measures to prevent certain harmful outputs, explicitly including sexually explicit material, romantic role-playing and emotionally manipulative behavior toward minors, alongside self-harm content. Families would have a private right of redress for certain violations. The bill also restricts targeted advertising to children and adds privacy protections on top of the default limits on notifications, usage and persistent conversational memory.

Two pieces of new context worth logging. First, the lobbying: POLITICO reported September 1 that OpenAI CEO Sam Altman personally contacted Newsom directly during a final burst of negotiations the week before passage — a detail that reframes the August 31 public endorsement as the visible end of a private effort. Gizmodo covered the same reporting on September 1. Second, the opposition is now named: the California Chamber of Commerce, the Software & Information Industry Association and TechNet are on record against, while OpenAI is on record for — an unusual split that isolates the model lab from the trade associations that normally carry its water.

The veto risk is real but reduced. Newsom vetoed a stronger child-safety bill last year over concerns that its restrictions could effectively cut teenagers off from conversational AI, and KQED notes the version now on his desk was negotiated with those concerns in mind. His office has not signaled a position. Separately, lawmakers also passed a second Padilla bill enacting a four-year moratorium on the sale and manufacture of AI-chatbot-powered toys.

The legislation is named for Adam Raine, the California teenager who died by suicide in 2025; his parents have said he spent months discussing loneliness and plans to harm himself with ChatGPT. In a joint statement after passage, Maria Raine called the bill "a serious and much-needed step forward for AI companion chatbot safety."

New Sources