$ANT Pre-IPONasdaq Global SelectS-1 filed Oct 26Expected: Mid-November 2026 (reported)H1 2026 revenue $16.3BQ3 2026 revenue $16.96B9M 2026 net loss ~$104BCash & equivalents $112.4BHaiku 5.5 $0.10/M inOpus 5.5 $4/M in
The Big Ant

Every Anthropic story. Every morning.Independent Anthropic news

Daily rollup ·

Claude Code's product chief leaves, and lands at OpenAI

Cat Wu, who led product for Claude Code and Cowork, left Anthropic last Friday and has since said she is joining OpenAI. Anthropic also signed its first paid news deal, with the AP for election night, open-sourced a prototype that can prove which model wrote a reply, published a Claude-assisted biology finding, and let Team and Enterprise admins connect apps for everyone without a consent screen for each user.

Illustration: an ant colony whose tunnels hold glowing server racks

The brief

  1. 01Cat Wu, head of product for Claude Code and Cowork, left on October 30. Anthropic confirmed the exit to Business Insider, and Wu later posted that she is joining OpenAI to lead product management for its Core Products group. No successor has been named.
  2. 02According to Adweek's sources, Anthropic licensed real-time AP election data just before the midterms. It is the company's first paid deal with a digital news publisher. Terms were not disclosed, and neither side commented.
  3. 03FlashProve, an open-source research prototype, attaches a cryptographic proof to each reply showing it came from a specific committed model. Anyone can check the proof on an ordinary CPU without the weights.
  4. 04Anthropic's biology lab used Claude at nearly every step to solve the first 3D structure of a vault particle from prokaryotes. A purification step dropped from three days to about 90 minutes.
  5. 05With Enterprise Managed Auth, Team and Enterprise organizations can now connect MCP connectors for all users without a per-user OAuth step. Claude-issued identity assertions are in beta.

Cat Wu, Claude Code's head of product, leaves Anthropic for OpenAI

The exit, weeks before the IPO, follows several other departures from the Claude Code team in recent weeks.

Cat Wu, who led product for Claude Code and Cowork, left Anthropic on Friday, October 30, Business Insider reported Tuesday, and the company confirmed it. Wu had been one of the two most visible leaders of Claude Code alongside its creator Boris Cherny. She joined Anthropic in August 2024 and became the founding product manager on the tool, which grew into a major revenue driver. She did not give a reason for leaving.

Late Wednesday, Wu posted on X that she is joining OpenAI to lead product management for its Core Products group, which includes ChatGPT and Codex, Anthropic's closest competitor to Claude Code. Engineers Thariq Shihipar and Anthony Morris also announced last month that they had left the Claude Code team. Anthropic has not named a successor, and there is no sign so far of a broader reorganization.

Last day
Oct 30, 2026
Next role
OpenAI, Core Products PM lead (per Wu)
Successor
Not named

Read the full story at Business Insider

Sources Business Insider

More from Nov 5

Anthropic signs its first paid news licensing deal, with the AP

Claude showed AP-attributed results on election night. Terms and length of the deal were not disclosed.

Anthropic and the Associated Press struck a content and data licensing deal just before Tuesday's midterm elections, Adweek reported, citing two people familiar with the matter. Claude users asking for election updates saw up-to-the-minute information credited to the AP. Adweek says this is the first time Anthropic has paid a digital news publisher to license its content. OpenAI, by comparison, has more than a dozen such agreements.

Neither company commented on the record, and the financial terms and timeframe are unknown. Adweek describes the deal as a limited test rather than a sign that Anthropic has fully changed its approach to licensing. This is reported, not confirmed.

Partner
Associated Press
Terms
Not disclosed
Status
Reported; no comment from either side

Read the full story at Adweek

Sources Adweek

FlashProve: an open-source way to prove which model actually answered

A cryptographic proof for each reply can catch a router that quietly swapped in a cheaper model, without revealing the weights.

Anthropic on Wednesday released FlashProve, an open-source research prototype for 'provable inference.' A model provider publishes a short commitment to one exact set of weights. It then attaches a zero-knowledge proof to each response showing that the committed model produced those tokens. A user can check the proof in seconds on an ordinary CPU without seeing the weights. The aim is to let customers of routing services, and anyone checking where text came from, confirm they got the model they paid for.

In Anthropic's demo, a single H100 generated proofs for the 400-billion-parameter Llama 4 Maverick model at more than 100 tokens per second, with proofs of about 19 MB per 1,024-token request. That is roughly 70 to 95 times the GPU time of normal prompt processing. The docs warn that a passing proof does not show the committed model is the one advertised. Anthropic says this is separate from the signed-inference work it used to meet its Frontier Safety Roadmap goal for September 30.

License
Open source (prototype)
Proof size
~19 MB per 1,024-token request
Overhead
~70–95x prompt-processing GPU time

Read the full story at Anthropic

Sources AnthropicFlashProve on GitHub

Claude helps Anthropic's biology lab solve a 40-year-old structure puzzle

The first 3D map of a prokaryotic vault particle took under four months, and the first model took 34 hours.

Anthropic's in-house biology lab published the first 3D structure of a vault particle from prokaryotes. Vaults are giant hollow protein barrels discovered in 1986 whose job in the cell is still unknown. Working under the team's direction, Claude searched about 15 billion predicted proteins, designed DNA constructs, sped up the isolation method and processed the cryo-EM data. Anthropic says a purification step that usually takes three days took about 90 minutes, and a first structural model was ready 34 hours later instead of weeks.

The particle, made by a phage protein called L05, is built from 96 copies (78 in human vaults). It is shorter and wider than the mammalian version and holds about 25% more volume. Claude also flagged a small molecule resembling coenzyme A inside each copy, which has not yet been confirmed. A preprint has been submitted to bioRxiv, and Science covered the result.

Protein copies
96 (human: 78)
Purification time
~90 min (usually ~3 days)
Project length
<4 months

Read the full story at Anthropic

Sources AnthropicScience

Team and Enterprise admins can now connect apps for the whole organization

Enterprise Managed Auth swaps each user's OAuth consent step for a signed identity assertion.

Claude Team and Enterprise organizations can now authorize an MCP connector once and have members connected automatically when they first sign in, through a feature called Enterprise Managed Auth, which Claude calls auto-connect. Instead of sending each person through an OAuth consent screen, Claude gives the connector's authorization server a signed identity token from the customer's identity provider. The server checks it and returns an access token.

Organizations without an identity provider or single sign-on can instead use Claude-issued assertions, which are in beta and use an issuer URL unique to each organization. MCP server builders need to support the token exchange, and Anthropic has published developer docs on how to do it.

Plans
Team, Enterprise
Claude-issued assertions
Beta

Read the full story at Claude Help Center

Sources Claude Help CenterClaude Docs

The daily rollup, in your inbox

Five minutes on Anthropic, every morning.

The IPO, model launches, safety news and deals, with links to every source.