Anthropic weekly, September 14–21: One Claude, a $100B run rate, and an antitrust suit

Anthropic spent the week folding Claude into a single product surface, publishing figures on how much of its own research Claude now runs, and drawing both a reported $100 billion annual revenue rate and an antitrust suit over the slowdown its chief executive asked the industry to accept.

On September 14 Anthropic launched Claude for Financial Advisors, connecting Claude to portfolio analytics and wealth-management software from BlackRock, Charles Schwab and Addepar, along with Envestnet, iCapital, Orion, Wealthbox, Wealth.com and Zocks. The toolset covers meeting preparation, portfolio review and follow-up work 1.

Two days later Anthropic merged Claude chat and Cowork into what it calls one Claude, folded Claude Design into the same interface, and introduced Claude Docs and Slides in beta. Documents export to Google Docs or Microsoft Word, presentations to PowerPoint or PDF 2. The merged interface reaches Pro and Max subscribers on web, desktop and mobile first, with Team and Free plans later 3.

Claude Code's Projects feature returned on September 17 with a coordinator directing threads that each run as a cloud session on their own copy of the repository, resolving overlapping edits as an ordinary merge conflict. Beta access opened to select Pro and Max subscribers 4.

The developer platform shipped two changes: the Messages API can now compact a conversation on demand, returning a signed summary block that replaces earlier turns on later requests, and the Compliance API's local session endpoints began returning Claude in Chrome transcripts for Claude Enterprise organizations 5.

Anthropic's new R&D Automation Index measures how much of its own research and development Claude performs. As of August, Claude "leads" 26% of that work, more than 90% of it sits at or above "AI collaborates," and no measured slice is fully autonomous. The rating scale comes from Epoch AI, an independent nonprofit, and the share at "leads" stood under 1% in February 67. The index rests on roughly 15,000 July 2026 tasks rated by Claude research agents, with company staff rating their own areas blind as a check 6.

The same post disclosed agent oversight: about 30,000 agents doing research and engineering work on its main internal platform at any one time in August, every action screened before execution, and about one in 47,000 of more than a billion monthly decisions blocked. Roughly 6% of the compute Anthropic used for AI research went to safety work in a sampled week in July, rising to 12% for AI-run research — figures the company calls conservative 6.

On September 18 Anthropic named its first embedded evaluator: Faculty, the AI division of Accenture, will red-team models, run alignment assessments and test safeguards from inside the company, with each side expecting to invest at least $1 billion over five years 89. Accenture shares rose 8% after hours 10. Anthropic called the arrangement non-exclusive, said it is in talks with METR about piloting embedded evaluation on nonprofit funding, and acknowledged that standards for evaluator access do not yet exist 8.

Novo Nordisk said on September 16 that it will test Anthropic's models and Claude Science in its research and development workflows. The Danish drugmaker already uses Claude to automate trial report generation, cutting patient-documentation timelines from months to minutes 11.

A day later Anthropic opened applications for its Life Sciences Verification Program, granting vetted organizations access to Mythos, Opus and Sonnet models through classifiers tuned to allow more biology work 12. Standard Use grants cover a whole team for daily work and renew annually. High-risk Use grants attach to a single project, renew every six months, and lift the biology blocks; they are available for Claude Opus 5 and Sonnet 5, with Mythos awaiting agreement with the US government. Enforcement shifts from blocking each request to monitoring traffic offline against the use cases an organization declares when it applies, requiring 30-day retention of flagged traffic. It launches in beta for teams and institutions; individual plans follow later 12.

Reuters and TechCrunch then confirmed the wet lab behind that program. Anthropic's head of life sciences, Eric Kauderer-Abrams, said the Bay Area facility does physical biology work, and a spokesperson said its focus is fundamental biology rather than drug discovery specifically 1314.

Anthropic's annualized revenue is expected to top $100 billion this year, Bloomberg reported on September 18, attributing the figure to the New York Times and people familiar with the matter. The run rate stood at $65 billion at the end of July and about $9 billion at the end of 2025. Anthropic declined to comment 15.

Reuters reported the next day that Anthropic is weighing a new model release ahead of the listing, partly to answer OpenAI's GPT-6 Astra. Astra accounted for about 13% of enterprise AI spending tracked by the expense platform Ramp, against about 8% for Claude Fable, and OpenRouter users spent more on OpenAI models than on Anthropic's for the first time in more than two years. The same sources said the IPO could slip past the November 3 midterm elections, with marketing at mid-October at the earliest 16.

Anthropic also joined Google, Nvidia and grid-software company Emerald AI in a coalition aiming to free 100 gigawatts of grid capacity by pausing data-center work at peak times 17.

China's state-backed Global Times called Dario Amodei's essay a "Cold War playbook" aimed at excluding China from global AI governance. President Trump, asked about calls to slow development, said "very negative forces" were raising exaggerated concerns 18.

On September 15 OpenAI's global policy chief, Chris Lehane, confirmed to reporters that OpenAI has worked with Anthropic and Google DeepMind on AI safety for weeks. He said the companies do not need an antitrust waiver, and OpenAI backed a provision in the FRONTIER Act that would place independent verification organizations inside frontier labs 19.

Microsoft's AI chief, Mustafa Suleyman, argued on September 16 that Anthropic made a mistake by embedding speculation about consciousness and model welfare in Claude's training materials. Suleyman said a model's statements about its own possible feelings reflect that training, and that teaching a system it may deserve welfare makes it harder to switch off 20.

Four plaintiffs sued Anthropic, OpenAI, SpaceXAI and Google on September 18 in the Northern District of California, alleging that the public calls to pace frontier development amount to an illegal agreement among competitors. Nick Rowley, one of the lawyers who filed the case, said the safeguards should be set by government with public accountability; the plaintiffs signalled they intend to seek class-action status 21.

The period recorded no leadership change, no new equity round, and no ruling in Anthropic's pending litigation.