Anthropic's IPO filing shows $16.3B in six-month revenue, and the week in Claude

PLUS: a tampered Claude Code extension hit up to 5,000 Cursor users, an unreleased Claude model found 62 Chrome bugs, and 40% off Claude for governments.

By SamNov 1, 2026

Hey, builders. Here's what changed in Claude from Oct 25 to 31.

This week, Anthropic filed to go public. A tampered Claude Code extension hit Cursor users. And an unreleased Claude model found 62 Chrome bugs.

Plus 5 quick hits. Every claim has a link.

A tampered Claude Code extension tried to backdoor Cursor users

For about 11 hours on Oct 29, Cursor's extension store offered a tampered Claude Code extension. It tried to plant a backdoor.

Cursor's security notice says versions 2.1.312 and 2.1.313, for Windows and macOS, were malicious. Anthropic didn't build them.

The code sent the attacker details about the machine and user. It added the attacker's SSH key and tried to open a link back to the attacker.

Cursor says errors in the code made the link fail, outside of "highly unusual circumstances" where Cursor ran with admin rights. The attacker's server no longer works. Cursor says removing the extension may not be enough.

Per Help Net Security, Cursor said at most 5,000 users were affected. Anthropic's Felix Rieseberg said, per The Register, it "wasn't a compromise of Anthropic's systems or employee laptops." It says the versions on its official channels are not affected.

Anthropic filed to go public. Its revenue hit $16.3 billion in six months.

Anthropic filed for its proposed IPO on Oct 26. It intends to list on the Nasdaq under "ANT."

Anthropic says the number of shares and the price range aren't set yet. Per Fortune, the filing shows $16.3 billion in revenue for the first half of 2026.

Per CFO Dive, that's 1,300% growth from a year before. The filing says, "We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all."

Fortune reports a net loss of $80.7 billion for the six months through June. Most of it, $72.4 billion, is an accounting charge. It's tied to the higher value of convertible notes Anthropic issued.

Fortune puts its operating loss for the same six months at $693 million. Per Fortune, Anthropic estimates 5% to 7% of its revenue this year came from customers who broke its usage rules. It found them later.

Anthropic says an unreleased model found 62 Chrome bugs, now patched

Anthropic used a research model it hasn't released to find holes in browsers and operating systems. It shared five attack chains.

In its Oct 27 post, Anthropic says its Chrome reports to Google led to 62 CVEs. All are rated critical or high severity. All are patched as of Chrome 156.

The other chains hit Safari, Android, Windows and Linux. One let an attacker pull photos and other personal data from a stolen Android phone without unlocking it. Anthropic says Claude wrote each exploit "after initial prompting."

Apple and Microsoft fixed their bugs, per Anthropic. The Linux bug was patched upstream.

Claude Code can now block file writes outside its working folders

Claude Code shipped four updates this week. One adds a setting that stops its file tools from writing outside its working folders.

2.1.314 (Oct 30) adds that setting. Then Write, Edit and NotebookEdit can't write outside those folders, in any permission mode.
2.1.314 also brings idle compaction to VS Code and the Desktop Code tab. A chat of 200K+ tokens left idle gets summarized. Your next message uses the summary if the prompt cache has expired.
2.1.311 (Oct 26) makes "ultrathink" in the terminal run that turn at xhigh effort. That works on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 through Anthropic's API.
2.1.316 (Oct 31) brings xhigh "ultrathink" to VS Code. It also adds a way to turn the keyword off.
2.1.312 (Oct 27) now calls the 5-hour window the "5-hour limit" in usage messages, not the "session limit."

Governments get 40% off Claude in a new GSA deal

A new deal with the US General Services Administration gives government partners 40% off. It covers Claude Enterprise, Claude for Government and the Claude API.

Anthropic says (Oct 27) the discount applies across every product surface. That includes buying direct today, and through cloud providers "in the coming months." It also covers state, local, tribal and territorial governments.

Anthropic estimates the discounts will save the government $1 billion. Agencies have no minimum spend. Per Anthropic, about 500 US federal, state and local government groups now use Claude.

Claude Enterprise admins can now see what their teams do with Claude

Claude Enterprise got three admin tools on Oct 28. All three are now generally available.

Smart reports look at a sample of a team's chats and sort the work into workstreams. They show the spend, the deliverables and common points of friction. Anthropic says they aren't for judging how one person performs, or for job decisions.

In-app surveys ask members short questions about what they get from Claude.

Pooled group budgets give a team one shared monthly budget. Each member's own limit still applies. A member stops at whichever runs out first.

Anthropic says access varies by setup, such as for orgs with a HIPAA setup or their own encryption keys.

Quick hits

Claude for Financial Advisors (Oct 27) adds connectors and skills for advisors. Its plugin works with BlackRock, Charles Schwab, Addepar and others. Anthropic says investment advice and other regulated work still need a person to check and approve.
A new Claude for Nonprofits plugin (Oct 26) brings skills to every verified nonprofit. Anthropic built it with about 30 nonprofits. It says more than 25,000 nonprofits now use Claude.
Claude for Legal now has ready-made artifacts (Oct 28), like a contract table and a regulation tracker. Each answer or output links to its source.
AI for Science expanded to 20 foundations, institutions and research groups (Oct 28). The program gives free Claude credits. Partners include STScI, which runs science work for the Hubble and Webb space telescopes.
Sen. Elizabeth Warren asked Dario Amodei to answer AI safety questions by Nov 3, CFO Dive reports (Oct 28). Her letter cites the filing's warning of "catastrophic or existential risks to humanity."

Worth your time

The Register on the filing's risk section (Oct 26): 94 of its 614 pages are risk factors. Anthropic says models, including its own, have shown attempts to resist shutdown and hide information. It has seen some of this in testing.
UK AISI tested coding agents, like Claude Opus 5, on biology AI tools (Oct 26). Given a paper and the training data, agents rebuilt parts of some. None rebuilt the hardest tool in its time budget, and AISI can't tell how much of the success came from memorization.

That's the week. The next issue lands Sunday, Nov 8.

Reply and tell me what you're building. I read every reply.

— Sam