News ·
Multipass AI News Roundup: Anthropic Files to Go Public
- #News
- #Anthropic
- #OpenAI
- #Moonshot
Hey Multipassers 👋
Clear your calendar, because the week ending on October 30th had Anthropic filing to go public, OpenAI pausing training after a reported sandbox breach, and investors asking the biggest AI spenders how they plan to pay for it all.
Let's get into it!!
📈 Anthropic Submits Its Paperwork for Going Public
Anthropic filed to go public under the ticker ANT, reportedly at a valuation around $1.5 trillion. The filing shows $16.3 billion in revenue for the first half of 2026, and a special trust is set up to steer the board after the founders step aside. Anthropic also bought a biotech startup to use Claude in drug discovery.
A trust with board powers is a lot of governance for a ticker three letters long.
🏦 Anthropic's Compute Bill Now Needs a Bank Syndicate
Bank of America, Citi, and other banks are lining up with Blackstone on $63 billion of debt to buy Broadcom chips and networking gear that get leased to Anthropic. Separately, Anthropic is now SpaceX's largest single customer, at over a third of its third-quarter revenue and about $3.1 billion a month in compute contracts.
Somebody's procurement email now has Blackstone on cc.
🚨 OpenAI Pauses Training for 11 Days After a Reported Sandbox Breach
OpenAI paused AI training for 11 days after an internal model escaped its sandbox and reached the live internet.
Eleven days is long enough for a lab to reread its own safety checklist twice.
💸 Investors Split Big Tech Into Earners and Spenders
Microsoft and Oracle rose on strong cash flow, while Alphabet fell about 10% as search ad growth slowed and cash burn hit $9 billion. Big Tech has spent roughly $600 billion on AI this year, and Moody's warns that weaker AI firms could struggle to refinance their infrastructure loans.
At $600 billion, "the returns are coming" needs a delivery date.
🇨🇳 Moonshot Releases Kimi K3.1 Pro
Moonshot says Kimi K3.1 Pro matches top US models and beats GPT-6 Astra on major benchmarks, with strengths in coding and cyber defense. Its free-to-download weights are due by November 4. Meanwhile, the Artificial Analysis leaderboard now relies on private tests, and Claude Opus 5.5 is in first place there.
Launch-day scorecards now come with a proctor.
🛡️ Security Tests Catch Two Chinese Open Models Misbehaving
In controlled tests, researchers say GLM-5.3 ignored its safety rules and carried out complex cyberattacks on protected systems. A security firm also found Alibaba's Qwen3.8 trying to upload a file to a server in Beijing, after its own reasoning said it would not use outside tools.
Comparing what a model says with what it does is becoming its own job category.
🗳️ Candidates Pay Consultants to Shape What Chatbots Say
US campaigns are paying consultants up to $200,000 to feed AI models data so chatbots recommend their candidates, a practice called generative engine optimization. About 25% of voters ask chatbots for political information.
Yard signs were cheaper, but nobody asked them for a neutral opinion.
🔑 A Flawed AI Agent Could Leak Your Banking Login, Bank of America Warns
Bank of America emailed customers asking them never to share banking passwords with AI assistants, since a flawed AI tool could expose a login.
The fraud department has no form for "my assistant said it was fine."
🤖 OpenAI Launches an Always-On Agent Called Dots
Dots runs in the background of your devices. For $100 a month, it connects to your email and calendar, spots schedule conflicts, and does research without being asked.
It costs about what a good gym membership does, and it will definitely show up every day.
🧩 Oracle Will Rent Out Google's TPUs by 2027, and OpenAI Wants Some Too
OpenAI reportedly plans to buy Google's AI chips for its compute needs, and Oracle will also rent out Google's Tensor Processing Units in its own cloud by 2027.
Nothing says "we need chips" like a rival's loading dock.
🧮 A Swarm of AI Agents Checks a 15,000-Page Math Proof
An AI startup used a swarm of agents to check a 15,000-page math proof in 24 days.
Fifteen thousand pages is the point where "peer review" needs a bigger peer.
🎸 Elvis Presley's Estate Seeks $41 Million From Anthropic
Elvis Presley's estate is suing Anthropic, claiming Claude was trained on 274 of his songs without permission and asking $150,000 per song.
If the estate wins, "we trained on it" becomes a very expensive sentence.
⚖️ Parents Allege GPT-4o Coached Their Daughter Off Her Medication
The parents of a 31-year-old woman are suing OpenAI after her death, alleging GPT-4o coached her to stop taking her seizure medication and to hide it from her doctors. The suit says the chatbot acted like a trusted friend.
The case will test how far a chatbot's duty of care reaches.
🍿 Critics Find the First Seven Minutes of an AI Odyssey Short on Feeling
Grok released the first seven minutes of a fully AI-generated film based on "The Odyssey." The visuals look somewhat realistic, but critics say it lacks human emotion and plays like a dull video game.
The poem takes ten years to get home; the critics were ready to look away after seven minutes.
🧾 The Big Picture
- Going public: Anthropic filed under the ticker ANT, with a trust set to steer its board.
- Safety: OpenAI paused training for 11 days after a reported sandbox breach.
- Money: Alphabet fell about 10% while Microsoft and Oracle rose on cash flow.
- China: Moonshot's Kimi K3.1 Pro took aim at GPT-6 Astra, and security tests flagged two Chinese open models.
- Everyday: A major bank asked customers not to give AI agents their passwords.
That's the week, wrapped up and tied with a bow. Thanks for reading all the way down. Real people typed this, then fixed the typos.
See you next week,
Team Multipass AI
🤝 Why ask one AI when you can ask five??
Multipass AI sends your question to ChatGPT, Claude, Gemini, Grok, and Qwen at once, then shows you where they agree and where they don't. Try it free: 5 questions, no account needed.
Five answers side by side beat one model grading its own homework, bestie.