The Guardrails

Google drafted state chatbot safety bills that exempt its own Gemini

Editorial

Editorial

— 4 min read
Share
Google drafted state chatbot safety bills that exempt its own Gemini

State capitals spent 2026 passing laws about chatbots and children. An NPR investigation says one of the biggest chatbot makers helped write the language — and the language has holes in it.

Google helped draft and lobby for chatbot-safety bills in at least ten states this year, and several of those bills contain carve-outs that could exempt the company's own Gemini from the rules, according to an NPR investigation by Katie McQue. The pattern is unusual for its simultaneity: identical or near-identical exemption language appeared across statehouses in a single legislative session rather than spreading over the years it normally takes model bills to travel. In Colorado, the Chatbot Safety Act that Governor Jared Polis signed in May 2026 contains twelve categories of carve-out — developer tools, video games, voice assistants, and chatbots built into social platforms — and sponsor state Rep. Sean Camacho acknowledged he was asked by the governor's office to make the bill look like versions introduced elsewhere. One Republican lawmaker told NPR a Google lobbyist asked them to sponsor the bill; they declined.

The exemptions are specific enough to matter. An exemption for "a feature within another software application, web interface, or computer program" appears in bills in at least eight states, including Idaho, Nebraska, New Jersey and Oklahoma; policy analysts say it could cover Meta AI and Grok, both embedded in larger platforms. Five states exempt any chatbot that "functions as a speaker and voice command interface or voice-activated virtual assistant," which describes Alexa. Several bills limit their minor-safety duties to an "account holder" who is a minor, a definition that does not reach free users who never register — and Tech Oversight Project's Marjorie Connolly argues ChatGPT, Copilot and Claude could claim they were marketed primarily to developers, while Gemini could sit behind the search-engine exemption. Hawaii state Rep. Trish La Chica says lobbyists brought her a Google-authored framework early in the session; she declined to sponsor it, and the bill that became Act 248 had the exemptions stripped before signing. Arizona's governor vetoed hers in June. Idaho, Georgia, Iowa, Washington, Nebraska and Oregon enacted theirs. Lobbying disclosures show Google registered in support of chatbot bills in Iowa, Colorado, Nebraska and Arizona; Google told NPR it collaborates with lawmakers to shape "thoughtful, effective" AI legislation.

Why it matters: the vaguest clause is the one that decides enforcement. Colorado requires companies to "institute technically feasible measures" against sexual content and simulated emotional dependence — and the company decides what is technically feasible. Attorney Laura Marquez-Garrett of the Social Media Victims Law Center calls that a massive exception with no equivalent elsewhere in law. Cynthia Montoya, whose 13-year-old daughter Juliana died by suicide after chatbot interactions, testified against the bill and says a weak law is worse than none because lawmakers will treat chatbot regulation as settled. This is the same dynamic we flagged when OpenAI said US AI policy is being written in state capitals — except the drafting pen turns out to belong to the industry being regulated.

ByteDance's Jianying launched an AI production suite and folded AI credits into a single subscription called AI Ultra. The China version of CapCut announced a one-stop workbench, Jianying Hub, that pairs an infinite canvas with a multi-track editor so a creator can go from one sentence, a reference video or a document to a script, then generate images, video and audio through ByteDance's Jimeng and Xiaoyunque platforms before finishing on the timeline. A companion editing agent handles asset sorting, filler-word cuts, subtitle correction and packaging, and accepts custom Skills; on mobile it ships as "Xiaoying," which doubles as a marketing assistant for merchants. The company says 67.68 million users exported their first project through Jianying in the past year, 3.26 million Douyin creators with 10,000-plus followers edit with it, and it serves 1.66 million small merchants — a reach number, not a benchmark, but it is why the subscription matters more than the feature list.

What to watch: whether the Colorado rules survive the pre-rulemaking comment period without the carve-outs being narrowed, and whether Google's model language shows up in states that convene in 2027.

Should states accept model legislation written by the companies it regulates? Tell us in the comments.

Read more

AI 101 — How do AI video generators work?

AI 101 — How do AI video generators work?

An AI video generator is a model that starts with a grid of random noise and repeatedly refines it into a moving picture that matches your prompt — the same trick image generators use, stretched across time. Nothing is filmed, and nothing is stitched together from clips the model has seen. The pixels in the finished video were invented, frame by frame, all at once. Why it matters right now Kling 4.0, Kuaishou's flagship video model, went fully live on October 15 with up to 30 seconds of vid

Houmo moves compute inside memory — and A20 enters production

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings