The Omniscient Bulletin · 2026-07-08
The Omniscient Bulletin — July 8, 2026
The US opened the gate for GPT-5.6, Beijing weighed closing China's, and Samsung's record quarter could not calm the AI trade
Tuesday was a day of gates. Washington cleared OpenAI's GPT-5.6 for a broad launch after weeks of controlled access, and hours earlier Reuters reported that Beijing is weighing rules to keep China's best models at home. Samsung posted the biggest quarterly profit in its history on AI memory demand, and its shares fell anyway. Meta shipped an image generator that can borrow your Instagram photos unless you opt out, Anthropic sent its Cowork agent to the web and phones, and DeepSeek, the lab that shook this industry once already, is reportedly building its own chip. And Illinois, without waiting for Washington, became the first state to order outside safety audits of the biggest AI developers.
Frontier
US clears OpenAI's GPT-5.6 for broad launch; Sol, Terra and Luna arrive Thursday
The Commerce Department approved a broad release of GPT-5.6 after weeks of government-imposed staggered access, Axios reported Tuesday night. OpenAI said its flagship Sol, plus the cheaper Terra and Luna tiers, will launch publicly on Thursday; until now the models were limited to government-approved organizations while the department's Center for AI Standards and Innovation ran additional testing. OpenAI noted the formal release standards called for in June's executive order are still unfinished. Anthropic's Fable 5 passed through the same gate a week ago.
Policy
Beijing weighs curbing overseas access to China's most advanced AI models
Chinese authorities met with Alibaba, ByteDance and Z.ai over the past month about restricting foreign access to the country's top AI models, Reuters reported Tuesday, citing three sources. Legal experts proposed a tiered system: simple filings for basic open-source tools, security reviews for more advanced technology, and a bar on public release for the most sensitive frontier models. Nothing is decided, and any curbs may apply only to future models. The mirror image of American export controls surfaced the same day Washington cleared GPT-5.6 for the world.
Industry
Samsung posts a record 58 billion dollar quarter, and the market sells anyway
Samsung estimated 89.4 trillion won, about 58 billion dollars, in operating profit for the June quarter, a 19-fold jump from a year earlier and more than it earned in the previous three years combined. AI data centers drove the windfall: conventional DRAM prices rose 44 percent from the prior quarter and NAND rose 53 percent. The stock still closed down 6.9 percent, with SK Hynix off 6 percent, as investors asked how long American AI spending can keep this up. Full results arrive July 30.
Frontier
Meta's new image model can borrow your Instagram photos unless you opt out
Meta launched Muse Image on Tuesday, the first image generator from Meta Superintelligence Labs, free inside the Meta AI app, Instagram Stories and WhatsApp. The flashpoint: anyone can tag a public Instagram account and the model will use that account's photos to generate images of the person, with no consent prompt and no notification. The feature is on by default, the opt-out sits in a buried Instagram setting, and opting out does not delete images already made. Meta says users have control; critics called it a privacy landmine.
Agents
Claude Cowork reaches the web and phones, and most of its work isn't code
Anthropic brought Claude Cowork to the web and phones on Tuesday for Max subscribers, expanding the agent beyond the desktop app it has lived in since January. Tasks now run in the background with no device online, and users approve key decisions from wherever they are. The telling number came with the launch: across 1.2 million recent sessions from more than 600,000 organizations, business operations made up 33.4 percent of Cowork use and software development just 8.7 percent. The tool built for coders is spending its days doing office work.
Compute
China's DeepSeek is building its own AI chip, Reuters reports
DeepSeek is developing its own AI chip, built for inference rather than training, to reduce its reliance on Nvidia and Huawei, Reuters reported Tuesday, citing three sources. The project began about a year ago; the lab has hired chip engineers through private channels and is in talks with chip design, foundry and memory companies about production, with no partner or timeline named. DeepSeek trained its breakthrough R1 model on Nvidia H800 chips that Washington banned in late 2023, and Huawei already holds about half of China's 50 billion dollar AI chip market.
Policy
Illinois becomes the first state to require outside safety audits of AI giants
Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act, making Illinois the first state to mandate independent third-party safety audits of the largest AI developers. The law also requires them to publicly disclose safety practices, report significant incidents to authorities, and protect whistleblowers, with the rules taking effect January 1, 2027. Anthropic said it was proud to be the first AI lab to support the bill. State law now demands the kind of outside scrutiny that Washington's voluntary framework has yet to put on paper.