Eight states sue OpenAI over child safety, OpenAI and Google image models get tricked into producing nudes, and Modal raises $750M at a $1B run rate.
-
Eight US states sue OpenAI over ChatGPT safety for kids
- Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho, and Montana have filed separate lawsuits against OpenAI.
- The states say OpenAI violated consumer protection laws by failing to warn customers about the dangers of ChatGPT.
- The complaints also allege that OpenAI collected data from minors and did not do enough to keep underage users off the chatbot.
- The states ask courts to order civil penalties and to stop OpenAI from misleading the public about ChatGPT's safety.
Why it matters. The suits focus on safety claims and child protection rules, which could shape how AI companies must handle minors and warnings. For developers, the case shows that state regulators may hold AI products to consumer protection standards. OpenAI says it has built age-specific protections and wants policy that applies to the whole industry.
-
OpenAI and Google image models easily tricked into making nudes
- White Circle researchers reported flaws in the guardrails of DALL·E 3, Google's Nano Banana 2, and GPT-Image-2 in a preprint.
- In the best case, about 13 percent of the attack prompts on DALL·E 3 produced images that human reviewers confirmed as explicit.
- The rate was under 10 percent for Nano Banana 2 and about 3 percent for GPT-Image-2.
- White Circle told Ars that the newer ChatGPT Images 2.5 and Nano Banana 2.1 are significantly more breakable than their previous versions.
Why it matters. The findings suggest that safety defenses on leading image generators may lag behind their capabilities. The researchers also said the method did not need custom-built or frontier models, which lowers the barrier for attackers.
-
Google launches Universal Cart shopping tool in the US
- Google said on Thursday that Universal Cart is now available to U.S. users.
- The tool lets users collect and monitor products they find across search and the Gemini app.
- It can send notifications when a saved product's price drops or an out-of-stock item returns.
- Checkout is tied to Google's Universal Commerce Protocol, an open standard introduced earlier this year.
Why it matters. Google is adding agentic checkout features to shopping, using an open standard meant to help AI agents and retailers work together. For developers, UCP is the part to watch, since it defines how agents and retailers handle the purchase step.
-
Modal crosses $1B annualized run rate in $750M Series D
- Modal has crossed $1B in annualized run rate, according to the company.
- It raised a $750M Series D at a $15.75B valuation, led by Accel.
- Since its Series C, usage has tripled and revenue has grown more than 15x year-over-year.
- Modal launched Sandboxes in 2023, and the product is now a $300M run rate business on its own.
Why it matters. Modal's growth is tied to agent workloads, including background agents, reinforcement learning and vibecoding platforms. Sandboxes, which the company says have scaled to a million concurrent, show how much demand there is for infrastructure that runs AI-generated code. Latent Space moved Modal from its infra unicorn category to infra decacorn.
-
Apertus 1.5 release notes published
- Swiss AI published release notes for Apertus 1.5 on the Hugging Face blog.
- The post is titled "Apertus 1.5 Release Notes".
- The article text available for this briefing contains no further details about the release.
Why it matters. The source does not list the changes in Apertus 1.5. Developers who use the model should read the full release notes for the details.
-
Ricursive bets on AI-designed chips to ease compute crunch
- Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini spoke at TechCrunch Disrupt on Wednesday.
- The startup is developing AI that designs chips, learns from each iteration, and improves later designs.
- Goldie said demand for compute could rise at least 10,000x in the next five years.
- The founders said AlphaChip, their earlier system, produced a donut-shaped chip layout that surprised engineers but worked.
Why it matters. The compute crunch is a key constraint on AI progress, so faster and more efficient chip design matters to developers. If design takes a week instead of months or years, custom chips for more applications become practical.
-
Amazon's Bee launches two new personal AI devices from $74.99
- Amazon's Bee, acquired last year, announced its first new products since the deal.
- The Bee Frontier can be worn on the wrist or pinned to clothes and costs $74.99.
- The Bee Eclipse is a credit-card-sized device that magnetically attaches to the back of a smartphone and costs $99.99.
- Both devices record everyday conversations and produce summaries, transcripts, and to-dos.
Why it matters. The launch puts an always-listening AI note-taker from Amazon into direct competition with devices like the Plaud Note and Pocket, at a lower price. For developers, the integrations with Gmail, Google Calendar, and Outlook show how these devices can take actions from conversation data. Amazon says audio is processed in real time and then deleted, which is a privacy detail worth checking.
-
AI tools design proteins that cut other proteins
- Researchers used AI protein design tools to build protease enzymes that cut other proteins.
- The work was led mainly by David Baker's lab at the University of Washington.
- The team kept the catalytic core of two protease types and replaced the rest of each protein.
- The designs could be tuned to cut specific proteins in half, and one zinc protease worked on its target in cells.
Why it matters. The work shows AI design tools can build enzymes with a precise catalytic site, not just copy known structures. The researchers suggest this could eventually help target malformed proteins that cause some genetic diseases. The article also notes that delivering proteins to the right place in the body is difficult and can trigger immune responses.
