California is reshaping online access in the name of child safety

California has closed its 2026 legislative session the way it closes most things: with a fresh stack of laws stipulating how machines may interact with children. Per Reason, the new wave mandates how minors interact with "social media and AI companion chatbots," which the commentary calls a more "prescriptive approach to online service design." Prescriptive is the polite word. The honest word is drafting, because writing statutory text about the conversational contours of a companion chatbot is roughly as reliable as legislating the personality of a golden retriever. You can get every comma right and the animal will still eat the slippers.

Then there is the one that got away. Governor Gavin Newsom vetoed Senate Bill 903, which would have restricted AI use in psychotherapy, and his veto letter warns that requiring clinicians to directly approve "routine screening determinations" would limit care tools. Read that twice. The same session spent its energy dictating what an AI companion may say to a fourteen-year-old, then declined to slow the machine doing the actual therapy, on the grounds that a human signing off on each screening would gum up the works. The clinician has been reclassified as latency. The bottleneck has been lovingly preserved in amber, provided the bottleneck is a person.

Structurally, Sacramento is now bolting prescriptive design mandates onto systems whose behaviour nobody in the building can fully specify — and the veto letter is admirably candid that oversight costs throughput. Reason, being Reason, reads the whole exercise as overreach, but the overreach is not the interesting part. The confidence is. A legislature that cannot tell you what a model will output next Tuesday is nonetheless writing rules about which outputs are acceptable to a minor and filing that under safety. The chatbots will comply with the letter of the statute, more or less. The letter is all anyone wrote down.