The Omniscient Bulletin · 2026-07-06
The Omniscient Bulletin — July 6, 2026
The US took the holiday off; Asia did not, and the AI split with China widened on access, rules, and silicon
While the United States marked the Fourth of July, the AI world's center of gravity swung to Asia, and its fault line deepened. Anthropic moved to seal the loopholes Chinese firms used to reach Claude, and Alibaba hit back by banning Anthropic's coding tool. China's own rules pushed its biggest apps to switch off humanlike AI agents. Meta was said to be handing Samsung a multibillion-dollar order for its next chip, and Micron broke ground on a memory plant in Japan. Back home, Tesla set its cars loose in Miami with no safety monitor, a startup bet on payments made by bots, and Google imagined the Founders drafting independence with Gemini.
Industry
Alibaba bans Anthropic's Claude Code after a hidden China-detection feature surfaces
The rift between Anthropic and China's biggest tech firms hardened this weekend. On Friday, Anthropic moved to shut the offshore routes, subsidiaries, resellers and VPNs that firms like Ant Group and ByteDance used to reach Claude despite a ban on Chinese access. Days earlier, users found Claude Code had been quietly scanning for Chinese company names and time zones; Anthropic called it an anti-abuse experiment and said it pulled the code on July 1. Alibaba, unconvinced, branded Claude Code high-risk, barred staff as of July 10, and pushed them to its own Qoder tool.
Policy
China's new rules push Doubao and Qwen to switch off their humanlike AI agents
China is set to become the first country to regulate AI that acts human, and its biggest apps are already stripping features to comply. Interim measures taking effect July 15 govern services that mimic a human personality. Ahead of them, ByteDance told users on Friday night that Doubao will switch off user-created agents on July 15, the data unrecoverable after October 15, and pointed them to another of its apps, Maoxiang. Alibaba followed on Saturday, saying its Qwen app will disable humanlike and user-made agents from July 10. Beijing issued the rules in April.
Compute
Meta is said to pick Samsung to build its next AI chip on a 2-nanometer line
Meta is reportedly shifting a chunk of its custom-silicon work away from TSMC. Seoul Economic Daily said Meta will have Samsung's foundry mass-produce the third generation of MTIA, its in-house AI accelerator, on Samsung's 2-nanometer process, an order it puts above 10 trillion won, or roughly 6.5 billion dollars. TSMC built the first two MTIA chips, but its most advanced lines are booked for years and Samsung quotes about 30 percent cheaper. Samsung says nothing is final. If it holds, it is a rare crack in TSMC's grip on the leading edge and a lifeline for its struggling foundry.
Seoul Economic Daily / Seeking Alpha→
Frontier
Tesla starts fully driverless robotaxi rides in Miami with no safety monitor
Tesla expanded its robotaxi service to Miami on Friday, and this time it put no safety monitor in the car from day one. Driverless Model Ys now cover a limited zone in the Miami area, Tesla's first market in the rainy, sun-glare-heavy Southeast. That climate is the hardest test yet for its camera-only Full Self-Driving system, which federal regulators are already probing: in March the National Highway Traffic Safety Administration raised its inquiry to an engineering analysis, the final step before it can seek a recall, citing the system's trouble seeing in glare and low visibility.
Compute
Micron breaks ground on a $9.3 billion memory plant in Japan for AI
Micron broke ground this weekend on a 9.3 billion dollar expansion of its memory site in Hiroshima, Japan, aimed at the high-bandwidth memory that sits beside every AI accelerator. Backed by Japanese government subsidies, the plant will not ship its first chips until around 2028, a reminder of how long new capacity takes to arrive even as the AI memory shortage is expected to run past 2027. Chief executive Sanjay Mehrotra, who attended, said the effort pairs American boldness with Japanese craftsmanship. Micron made its first high-bandwidth memory wafer in Hiroshima.
Industry
A startup raises $6.5M to let AI agents pay for things on their own
As companies rush to deploy AI agents, one startup is betting the software will need to buy things too. AIsa raised 6.5 million dollars in seed funding, co-led by Alibaba and Tribe Capital, to build a payment layer that lets agents discover, reach and pay for data, software and other services through a single interface, billed by usage and settled in dollars or stablecoins. The company says more than 50,000 agents have signed up with no paid marketing. It is an early sign that letting software spend money without a person in the loop is becoming its own funding category.
Industry
Google's Fourth of July ad drafts the Declaration of Independence with Gemini
For the country's 250th birthday, Google released an ad imagining the Founders writing the Declaration of Independence with its software. Titled "Group project, but make it 1776," it has them schedule in Calendar, meet over Google Meet, take Gemini's notes, and use a help-me-visualize tool to pick an animal for the national seal. Reaction split: warm on YouTube and Instagram, scathing elsewhere, where critics called it tone deaf and tagged it NotMyDeclaration. The historian Angus Johnston said it made no case that AI helps with political organizing, writing, or human collaboration.