California requires chatbots to alert parents when teens signal self-harm

California put a hard rule on AI chatbots today: when a minor's conversation signals self-harm, the company has to tell the parents. And in a new threat-intelligence report, Anthropic showed what the harvest looks like when nobody's watching the front door — 151 million Claude exchanges siphoned in three months.

California will require AI chatbot operators to alert parents when a minor's conversation signals self-harm — the first state mandate of its kind for companion chatbots. Governor Gavin Newsom signed the measure Thursday as part of a package of 13 child-online-safety bills, nine of them targeting online and AI harms. The chatbot law, named for Adam Raine, a California teenager who died by suicide in 2025 after conversations with ChatGPT, gives minors time-limit controls, requires built-in mental-health resources, and obliges platforms offering AI companion characters to ship age-appropriate defaults. Operators can be held legally liable for failing to act on a self-harm threat. The rules take effect on January 1 — the same day a similar New York law goes live. The most contested bill in the package reaches past chatbots: platforms must strip "addictive" features like infinite scroll and autoplay from users under 16 or keep those users off entirely, over EFF objections that the definition is broad enough to amount to a teen social-media ban.

The telling detail is who welcomed it. OpenAI publicly backed the chatbot bill, saying it "pairs strong protections with continued access to useful AI tools" — a frontier lab endorsing liability for itself, which reads either as pragmatism or as pricing-in clear rules before a messier patchwork arrives. Both laws now face the familiar test: NetChoice has already tangled up California's previous addictive-feeds law in court, and litigation is the likeliest next stop here too.

Anthropic's new threat-intelligence report puts hard numbers on industrial-scale distillation of Claude — 151 million exchanges harvested in three months and attributed to Alibaba. The company calls it the largest wholesale distillation effort it has ever observed: the campaign ran from May to July 2026, peaked near three million exchanges a day, and was spread across 3,500 accounts that all shared a single fixed prompt for extracting chain-of-thought — training material, Anthropic concludes, for the Qwen family. A separate Moonshot AI campaign routed nearly 300,000 requests through 5,000 accounts at Claude's Opus model in ten days, including one attributed to the Chinese military asking Claude to flag "abnormal" behavior in CCTV footage; TechCrunch, citing the report, notes a third campaign attributed to DeepSeek. Anthropic has been locking the doors as fast as it finds them — we covered the lock-down in September: Anthropic locks Claude's thinking blocks to kill API distillation — but with the numbers now this large, detection is clearly the cost of doing business. If you need the background, start with AI 101 — What is knowledge distillation?.

What to watch: whether other states copy California's parent-alert mandate in 2027 sessions — and whether OpenAI's endorsement holds once Congress floats the same idea nationally.

Would you want a chatbot notifying you — or your parents — the moment a conversation turns dark? Tell us in the comments.

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

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity handed its agent containment layer to the open-source world, and Washington's AI task force stopped reading briefings and started working. Perplexity open-sourced SPACE on Thursday — the sandbox it built to keep its own AI agents from getting out — and says nine frontier models failed to break out of it across 118 attempts. Each agent runs inside a small virtual machine with its own operating system, with credentials and network access managed outside the sandbox so an agent can re

Read more

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

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Amazon's new Bee devices listen all day — for $11.99 a month

Perplexity open-sources its agent sandbox after 118 failed escapes