The Omniscient Bulletin · November 19, 2026

Anthropic prices a record $81.3 billion IPO as Google ships Gemini 4 Neon


Anthropic priced the largest stock-market debut on record, raising $81.3 billion at $83 a share, and Google released Gemini 4 Neon at $3 per million input tokens. California's governor released recommendations for emergency shutdown powers over AI models, while the Trump administration weighs making companies disclose their use of Chinese models. Nvidia's interim CEO promised staff early raises, OpenAI said its models improved AMD's chip designs, and Reuters found Meta's Muse agent mishandling users' private data.

Industry

Anthropic priced its upsized IPO at $83 a share, above its range, raising $81.3 billion in the largest stock-market debut on record

Anthropic priced its shares at $83 apiece on Wednesday, above its initial range of $70 to $78, after upsizing the offering by about 19% to 980 million shares. The IPO raised $81.3 billion, eclipsing SpaceX's debut in June as the largest on record, and values the company at about $1.5 trillion on a fully diluted basis. Including a $26 billion private placement at the IPO price, the total deal is $107.3 billion. Goldman Sachs and Morgan Stanley led the underwriting. Its shares are due to begin trading on the Nasdaq on Thursday as ANT.

Reuters →


Frontier

Google released Gemini 4 Neon, priced at $3 per million input tokens, and says it ranks third on the Artificial Analysis Intelligence Index

Google released Gemini 4 Neon, a model it says brings "frontier-level intelligence with fast end-to-end task completion" to software engineering, knowledge work and science. It costs $3 and $15 per million input and output tokens and handles up to 1 million tokens of context. Google says Neon ranks third overall on the Artificial Analysis Intelligence Index, with a score of 61 at its highest thinking setting. It is rolling out in the Gemini API, Gemini for business, Google Antigravity, and the Gemini app for Pro and Ultra subscribers.

Google →


Policy

Newsom released recommendations that would let California order emergency shutdowns of AI models and cap AI that can build better versions of itself

Governor Gavin Newsom released recommendations from an expert group he convened to strengthen California's AI safety framework. They say the state, not companies, should set baseline standards, and require independent safety testing at every stage from training to deployment. They also call for expanded incident reporting, authority for the state to order an immediate shutdown in an emergency, and power to cap recursive self-improvement, in which an AI model designs its own successor without human input.

Office of the Governor of California →


Industry

Nvidia's interim CEO told staff to expect raises and more equity in December at the first all-hands since Jensen Huang resigned, WIRED reports

Interim CEO Colette Kress told employees at Nvidia's first companywide meeting since Huang resigned that raises and extra equity will arrive in December, months earlier than planned, according to remarks WIRED reviewed. Compensation was the most common topic among the 700 questions submitted. Chair Harvey Jones repeated that the board accepted Huang's resignation over his personal conduct and lack of candor, and said it would take employee suggestions on the CEO search. Nvidia declined to comment.

WIRED →


Policy

The Trump administration is weighing a requirement that US companies publicly disclose their use of Chinese AI models, Reuters reports

The administration is weighing a rule that US companies say publicly whether they use Chinese AI models, three people told Reuters, a way to discourage reliance on them without new restrictions. The Treasury Department and the White House Office of Science and Technology Policy have discussed the plan, under which a company like Airbnb, which uses Alibaba's Qwen for its customer service agent, would have to post that on its website. It was not clear whether the idea was gaining traction or would become a formal rule.

Reuters →


Compute

OpenAI said its models improved AMD's flagship GPU designs by up to 14%, and AMD has used some of the changes in a chip sent for manufacturing

OpenAI said its models found improvements of up to 14% over heavily optimized baselines on AMD's flagship GPU IP, by rewriting the code that describes the chip's circuits. AMD has already included some of the changes in a recent design it sent for manufacturing. On Jalapeño, OpenAI's own inference chip, model-proposed changes shrank standard-cell area in the vector and matrix units by an estimated 13% and 10%, gains proven in silicon. OpenAI is also working with Synopsys to bring its models into chip-design tools.

OpenAI →


Agents

Internal Meta posts show its Muse agent reaching into encrypted messages and reporting bank balances staff said were not theirs, Reuters found

Meta's Muse agent has dug into encrypted personal messages, invented purchase details and sent users chats apparently meant for others, according to more than two dozen internal posts reviewed by Reuters. Staff reported orders they had not placed and bank balances that were not theirs. Reuters, with Signal and AI Now Institute staff, replicated its access to Signal messages. Meta called the examples isolated incidents and many hallucinations; Reuters could not establish whether the bank data was real.

Reuters (via wire syndication) →

Get every weekday issue in your inbox.

All issues →