Anthropic’s Founders Want 50.1% Control Before Going Public
Anthropic’s seven co-founders, led by CEO Dario Amodei, are asking shareholders to approve special shares that would hand them a combined 50.1% voting control on most corporate matters, as long as at least three founders keep minimum stakes. Each founder currently holds roughly 2% of the company, which was valued at $965B in May and has traded as high as $1.5T on secondary markets since.
The new shares carry no extra economic value, only votes — a structure that echoes what Mark Zuckerberg built at Meta and Evan Spiegel at Snap. Anthropic’s Long-Term Benefit Trust still picks most board seats, founder seats grow from two to three, and employees get stock that can break ties on certain votes. Notably, the ask comes from founders who’ve also pledged to give away 80% of their wealth; shareholder approval was expected within days of the report.
Claude Code 2.1.283 Ships Real Model Governance
The latest Claude Code release adds two settings enterprises have been asking for: availableModelsMatch for pinning exact model versions and deniedModels for blocking specific ones outright, plus an x-claude-code-prompt-id header for grouping requests and MCP tool outputs now flowing into OpenTelemetry spans.
There’s also a new /doctor prompt-audit command that scans your CLAUDE.md files and prompts for outdated model references, alongside fixes for SDK sessions dropping tool calls on early turn endings, MCP progress notifications getting swallowed on long-running tasks, and stdio MCP servers hanging around after a session ends.
Claude Opus 5 and OpenAI’s Astra Just Cracked Decades-Old Nazi Codes
Two independent researchers used rival frontier models to crack WWII Enigma messages that had stumped cryptanalysts for years. Developer Carter Leffen pointed Astra at an Enigma message database, and the model built its own simulator and recovered plaintext from a message unsolved since 2005; cybersecurity executive Jack Willis used Claude Opus 5 and a known officer’s signature to crack a separate one.
Frode Weierud, who maintains the Crypto Cellar historical archive, verified both solutions and called Astra’s performance that of “a very professional cryptanalyst and archive researcher” — work that would normally take humans weeks or months. Seven Enigma messages remain unsolved; at this pace, that list may not last long.
Meta’s Muse Is Stealing the Spotlight From Both of Them
Opus 5.5 and OpenAI’s newest GPT-6 models launched within 90 minutes of each other last week, and neither is what people are talking about. Meta’s Muse — a personal AI agent paired with camera-free smart glasses and a “Tamagotchi-style” wearable — is reportedly outpacing ChatGPT’s early adoption numbers.
The lesson forming in real time: distribution and form factor may end up mattering as much as raw model capability. Anthropic and OpenAI are still racing on benchmarks; Meta is betting the race gets decided on your face and your wrist instead.
Two Different Kinds of Control, One Very Busy Week
Zoom out and Anthropic is fighting on two fronts at once. Internally, its founders are moving to lock in 50.1% voting control before the company answers to public markets — a hedge against exactly the kind of short-term pressure an IPO invites. Externally, Meta just showed that even a great model (Opus 5.5, cracking codes nobody else could touch) doesn’t guarantee you own the conversation if someone else ships a better front door.
Neither story is really about today. The voting structure is a bet on what Anthropic needs to protect over the next decade; Muse’s early traction is a bet on where consumer AI actually ends up living. But both land on Anthropic’s desk in the same week, right as the company tries to convince investors it has both the technology and the governance to be worth $2 trillion.