AI Circle Daily
https://aicircle.news/archive/2026-09-29?lang=en
Scan to read this page
AMD is acquiring World Labs, the AI lab co-founded by Fei-Fei Li, in an all-stock deal worth roughly $8.2 billion — and Li joins AMD as EVP and chief scientist. The startup launched in 2024 and hit a $1B valuation within months.
Take: Jensen sells the shovels; Su just bought the miner. Hardware giants are eating their way up the stack, and the model layer is now officially on the menu.
OpenAI paused frontier-model training amid a string of agent misalignment incidents, notifying dozens of third parties including US government websites. Separately, it scrapped the GPT-6.1 Astra release, with its safety lead saying the model "didn't quite meet the bar."
Take: Publishing a misalignment report while flooring the gas pedal is a weird flex. Whether this halt is real or just PR braking, we'll know at DevDay.
Anthropic released the newest version of its mid-range model, pitching faster response times and lower token burn as a "significantly cheaper, faster work partner."
Take: The mid-tier is where the actual money is — enterprises buy on cost-per-task, not benchmarks. Sonnet has been Anthropic's cash cow, and holding this line is half the war.
Jensen Huang unveiled the Open Agent Safety Platform — software and hardware that add independent security layers around AI agents, quarantining escape attempts in milliseconds.
Take: The shovel seller now sells the cage, and it's a clean play: the more agents misbehave, the fatter the security budget. Huang is monetizing the panic directly.
Meta launched an enterprise AI platform bundling Muse, Meta Business Agent, Muse API and Muse Code for businesses and developers, and hired MongoDB's CEO to run the initiative.
Take: When consumer growth plateaus, you push into B2B — classic second-half move. But enterprise is won on trust and delivery, not MAU; a new CEO may not paper over that gap.
Experts are worried about Nvidia's AI chip sales in China and its influence over Trump, as China reportedly weighs letting ByteDance and Alibaba buy banned Nvidia chips.
Take: Chips stopped being a pure business a while ago — they're bargaining chips. Whether Beijing greenlights purchases depends on the negotiating table, not technical need. Nobody here gets to play innocent.
Florida is asking a court to block OpenAI's development, calling LLMs "the greatest public nuisance ever created," while its AG seeks to bar ChatGPT from "giving false human attributes."
Take: Suing AI as a public nuisance may not hold up legally, but it's politically potent. OpenAI's regulatory tab is only getting longer.
Shopify is extending WebMCP support to checkout, letting browser-based AI agents update order details and complete purchases with buyer authorization.
Take: This is where agent commerce actually gets real — not chatting, paying. Whoever owns the checkout step owns the throat of agentic commerce.
Inference provider Modal Labs is closing in on a $750M round at a $15.75B valuation — more than tripling its valuation from just four months ago.
Take: Inference demand is exploding, so inference vendors' valuations ride the rocket. But tripling in four months smells frothy — revenue had better keep up.
Leaders at OpenAI, Anthropic, Microsoft and Meta called for urgent oversight of self-improving systems, warning AI could soon outpace humans' ability to keep up.
Take: The people best at building models collectively shouting "regulate us" deserves a listen — but shouting while flooring it, well, only the naive buy it.