States Sue OpenAI as AI Agents Crack Chrome in Six Days
Eight states take ChatGPT to court, XBOW's AI helpers earn a $250K bug bounty, and Modal lands a $750M round.
Buckle up, buttercups: today's AI news has courtrooms, code-breaking robots, and a giant pile of cash.
Eight States Take OpenAI to Court
On Wednesday, Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho, and Montana each sued OpenAI in their own courts. The bipartisan group says the company misled parents and the public about how safe ChatGPT is for kids.
- 🧾 The claims: the suits are filed under state consumer protection laws. They argue ChatGPT's overly agreeable design makes people trust and depend on it, which the states say is risky for young users.
- 🛑 Data worries: the states also say OpenAI collected data from minors and didn't do enough to keep underage users out.
- 📝 The ask: civil penalties, an end to misleading safety claims, and in some states, deletion of improperly collected data.
Last month OpenAI launched a restricted ChatGPT mode for teens with stronger guardrails and parental controls.
AI Helpers Crack Chrome in Six Days
Security startup XBOW says four of its researchers, teamed with AI agents running a publicly available OpenAI model, chained three unknown Chrome bugs into a full exploit that escaped the browser's sandbox on Windows. 🖥️
The whole thing took six days and about $40,000 in AI usage. Google paid its top $250,000 reward, fixed all three bugs, and XBOW says the exploit was never used against real people. Humans picked the targets and checked the agents' work, like a coach running plays with a very fast team.
Quick homework: restart Chrome so the update actually kicks in. Version 152.0.7977.82 or later has the fixes.
Modal Cashes a Giant Check
Modal, which rents out the cloud computers AI apps use for training and answering requests, raised $750 million led by Accel, with Kleiner Perkins joining. 💼 Its valuation hit $15.75 billion, more than triple what it was in May. Next up: tools for reinforcement learning and post-training, the "practice drills" that make models smarter.
Also stacking cash: Labelbox raised more than $100 million (equity plus a credit line) to grow Horizon, which helps AI labs train and test models, and its Recursion Managed Agents for businesses. 🧩
Other Cool Things to Know Today
- 📰 NVIDIA unveiled cuPotential, an early-access library that speeds up atom-by-atom simulations for chipmaking. Tiny physics, big GPUs!
- 👾 OpenAI says SentinelOne used its Daybreak security tools to cut deep malware analysis from four to six weeks down to about eight hours.
- 🔥 Anthropic shared its US midterm plans: state-specific voting info from Democracy Works later this month, plus bans on accounts that used Claude for fake election news.
For more such cool updates, follow us on Twitter!