THE DAILY TEN · 2026-10-15

Eight States Sue OpenAI Over ChatGPT Safety Claims and Risks to Children

plus 9 more stories

Also: StepFun Publishes Step 5 Preview Weights for Long-Running AI Agents; and Nvidia Releases Nemotron 3.5 Super VL and Bolt for Multimodal Agents. Every story links to its primary source.

Curated by Gaurav · October 15, 2026

01

Eight States Sue OpenAI Over ChatGPT Safety Claims and Risks to Children

Iowa Attorney General ↗

Eight states have filed separate lawsuits against OpenAI, alleging that ChatGPT harms young users and that the company misled families about its safety. Iowa announced its case alongside Kansas, Nebraska and West Virginia, with four other states also suing. The cases seek changes through state consumer protection laws; the accusations are allegations, rather than findings by a court.

02

StepFun Publishes Step 5 Preview Weights for Long-Running AI Agents

StepFun / Hugging Face ↗

The Chinese lab has published downloadable weights for Step 5 Preview, following its earlier hosted release. The model has about 600 billion total parameters, activates 27 billion per token and supports up to one million tokens of context for coding and professional work. Its model license includes additional commercial conditions, while the deployment guide calls for multiple high-memory GPUs.

03

Nvidia Releases Nemotron 3.5 Super VL and Bolt for Multimodal Agents

Artificial Analysis ↗

Nvidia has released Nemotron 3.5 Super VL alongside the smaller Nemotron 3.5 Bolt, both accepting vision and text inputs. Artificial Analysis reports that Super VL leads US open-weight models on its Intelligence Index, while using about 60% fewer output tokens per task than Nemotron 3 Super. The gains give developers another option for running their own agents, although stronger open-weight models from other countries still lead the index.

04

Anthropic Gives Six METR Evaluators Employee-Like Access to Its Safety Work

Anthropic ↗

Anthropic says six METR evaluators now have company hardware, office space and access to internal systems comparable to its own risk assessment teams. METR can publish its assessments without an Anthropic veto, subject to narrowly defined redactions that it can publicly challenge. The arrangement offers deeper outside scrutiny of frontier AI development, with customer data and certain other information excluded.

05

Google Rolls Out Universal Cart Across Search and Gemini in the US

Google ↗

Google is rolling out Universal Cart in the United States across Search and the Gemini app. Shoppers can save products from multiple retailers, receive price and restock alerts, find promo codes and compare payment perks. Checkout uses the Universal Commerce Protocol, with Google Pay or a handoff to the merchant, which remains responsible for the sale and customer support.

06

Grok Voice 3.0 Arrives at Eight Cents a Minute With Faster Responses

SpaceXAI ↗

Grok Voice 3.0 is available through the Voice Agent API at $0.08 per minute, matching its predecessor's price. SpaceXAI reports response latency below one second and a leading score on Artificial Analysis's speech-to-speech index. The company says the system already powers Tesla conversations and Starlink support, where roughly 70% of calls are resolved without a human; those deployment figures are company claims.

07

Amazon Launches Bee Frontier and Unveils Bee Eclipse for Personal AI Memory

Amazon ↗

Amazon has introduced the $74.99 Bee Frontier wearable and the $99.99 Bee Eclipse magnetic phone attachment to turn conversations into summaries, reminders and insights. Frontier is available now, while Eclipse is coming soon. Amazon says audio is deleted after processing, and optional connections to Alexa and Amazon shopping require Bee Pro, which costs $11.99 per month after the introductory trial.

08

New Robot-Arm Benchmark Shows Frontier AI Still Far Slower Than People

Artificial Analysis ↗

Artificial Analysis has launched an eight-task benchmark that lets coding agents control a real robot arm through cameras and an API. GPT-6 Astra leads the research preview at about 69 out of 100, but averages around 30 minutes and $16.36 per task. The evaluators say these systems remain roughly 100 times slower than people, exposing the gap between capable demonstrations and practical robot work.

09

Proof Launches x401 to Verify Who Authorized an AI Agent's Transaction

Proof ↗

Proof has launched x401, an open protocol that lets websites request verified identity and signed approval for a specific AI-agent action. A person reviews the transaction on their own device, and changing signed details such as the amount or recipient invalidates the evidence. The company also offers an identity wallet and an MCP server, giving businesses a way to check what a customer actually approved.

10

Prime Sandboxes Lets Developers Fork Agent Work From Saved Files

Prime Intellect ↗

Prime Sandboxes supports checkpoints that save a working filesystem and fork it into new sandboxes, allowing agent experiments to branch without repeating all their setup. The feature captures files rather than running processes or memory, so applications must restart after restoration. Documented checkpoint storage costs $0.10 per gibibyte per month, plus an upload charge, making retained experiment branches an explicit cost.

← 2026-10-14 All briefings

Tools in this briefing, reviewed in our directory:

GPT-6 Astra → Grok →

Track the tools behind the news:

→ AI Tool Leaderboard → Browse the directory