Everything That Happened in AI Today (Wednesday, November 11, 2026)

Google had AI generate more than six million lines of mathematical proof to check the encryption protecting private messages. Yes, six million.

Welcome to Around the Horn, where the AI industry apparently announces its next chapter all at once. One major lab is asking public-market investors to fund its ambitions, while its rival's latest financial results show just how expensive the race has become.

Meanwhile, researchers are using AI to check the mathematical foundations of secure messaging, and a stolen batch of AI credentials produced a six-figure bill.

The technology is getting more capable, and the invoices are keeping pace. New models, workplace tools, and creative apps are also worth a look. Let's get into it.

🆕 NEW From The Neuron

• Who's Investing in Anthropic's IPO? Seven tech giants have committed $26 billion alongside the offering. Our report follows their interests across Claude, cloud infrastructure, and the hardware it needs.
• Australia's AI Safety Plan examines proposed incident-reporting rules and penalties after an OpenAI agent announces a government data portal.
• NVIDIA's GR00T 2 asks whether learning from human video will help robots pick up new tasks faster, and which claims still need independent testing.

Around the Horn — Wednesday, November 11, 2026

Anthropic just took its IPO plans onto the road. The company behind Claude launched the roadshow for a proposed Nasdaq listing under the ticker ANT, with an expected share-price range of $70 to $78. Its preliminary offering includes 825 million shares: about 801.4 million sold by Anthropic and 23.6 million by existing investors. At the upper end of the range, the shares offered would have a gross value of roughly $64.4 billion before underwriting expenses.

For perspective, this is a proposed stock offering, not money already raised. Anthropic would receive proceeds from its own shares, while existing shareholders would receive the money from theirs. The company's SEC registration statement has not yet become effective, so the shares cannot be sold under the proposed offering yet. Investors will now weigh demand for Claude against the massive costs of developing models and operating AI infrastructure.

The roadshow also arrives during a crowded news cycle for the AI industry's finances. OpenAI reportedly generated about $10 billion in revenue in a single quarter while continuing to post large operating losses.

One company is asking investors to buy in, and the other is showing them what the race costs. The eventual IPO price, investor demand, and Anthropic's disclosures will tell us much more than the headline fundraising range alone.

🏆 TOP 5 NEWS (Around the Horn)

• OpenAI's revenue jumped, but so did the losses. Its third-quarter revenue rose nearly 50% to about $10 billion, thanks to its enterprise products and recent models. Operating losses remained enormous as investors examined the economics behind its growth.
• Google used AI to check Signal's encryption mathematically. DeepMind's AlphaProof Code verified important security properties of Signal's encryption software in four weeks. The analysis covered specified properties and code, not every part of Signal or every possible attack. Researchers also checked that the system rejected older versions with known bugs.
• Stolen AI keys helped generate a $740,000 bill. Railway's breach report says attackers obtained customer secrets that could have affected up to 100,000 workspaces, an estimate of maximum exposure rather than a count of confirmed victims. The company estimates that stolen credentials produced roughly $740,000 in unauthorized AI use.
• New York launched state-run AI testing. Governor Kathy Hochul introduced an independent AI evaluation program. Binghamton University's new research center will study model risks with researchers and state-backed computing resources, though access to unreleased systems remains a limitation.
• Google agreed to more safeguards for Spirit Airlines' employee data. Reuters reported that Google added de-identification and independent-review protections to its proposed $10 million purchase of business records from the defunct airline for AI training. Health information will be excluded, and the transaction still needs court approval.

Honorable Mentions

• AI exposure is worrying big investors. New York City's $325 billion pension system has passed on some AI investments because the portfolio is already heavily exposed to the sector. Reuters also found investors in AI-related debt are becoming more cautious as Congress prepares to scrutinize data centers and AI regulation.
• Australia proposed tougher AI incident rules. Australia plans an AI safety regulator and mandatory reporting of serious incidents after an OpenAI agent accessed a Medicare statistics portal. The government accepted 18 review recommendations, but the legislation has yet to be enacted.
• The US electricity forecast keeps climbing. The Energy Information Administration expects consumption to hit new records in 2026 and 2027, partly because of AI data centers. Its forecast rises from 4,216 billion kilowatt-hours in 2025 to 4,370 billion in 2027.

🍪 TOP TREATS TO TRY

• Gemini 4 Carbon: Google opened access to its frontier model for demanding coding, research, and multi-step agent tasks. Paid developers can use the API, while Google AI Ultra subscribers can try supported consumer tools. Google also previewed /swarm, which coordinates multiple coding agents.
• Businesses in ChatGPT: OpenAI introduced a beta Business Hub where companies can claim their profiles, update information, and view aggregate engagement insights. Eligible US ecommerce and online-service businesses can also create Business Agents to answer shopper questions.
• Claude's new workplace tools: Anthropic introduced Claude Sheets for spreadsheets and Whiteboards for collaborative ideas, while Claude Data adds shared dashboards. A Claude announcement emphasized interactive connectors that open apps such as Adobe and Atlassian beside chats. Claude Meetings is also entering beta for Team and Enterprise users, initially on Google Meet, where it can begin tasks during a call.
• Gemini for healthcare: Google Cloud previewed a healthcare-specific AI system that can consult records and coverage policies to help with tasks such as medical record summaries and insurance authorization reviews, subject to staff review.
• Gemini Robotics 2.5: Google DeepMind showed robots learning a broader range of physical tasks, including laundry and factory sorting. The company says partners are beginning to receive the new model; its reported factory performance is based on a controlled proof of concept.
• MiniMax H3-Image: MiniMax released downloadable model weights for image generation and editing. It claims 4K images in under 10 seconds, support for up to 16 reference images, and faster output than several competing systems.
• Qwen4-Max preview: Alibaba teased the Qwen4 family and made the Max variant available in preview through its Token Plan and selected products. The company's announcement emphasized coding and AI agents.
• ElevenLabs Scribe v3 and v3 Live: The new transcription models handle recorded and real-time audio. ElevenLabs says v3 made 27% fewer word errors than v2 in its multilingual tests, while the live version reduced meaning-changing errors by 43% against its real-time predecessor.
• Adobe's creative AI overhaul: TechCrunch covered new agents across Photoshop, Premiere, and Firefly, plus integrations that let outside agents work with editable Adobe project files. Creative Bloq detailed Photoshop 2027's new Relight and Generative Restyle features, along with tools for removing objects and changing image composition.
• Reddit Reviews: Reddit began testing AI-generated summaries of community opinions about products, with dedicated pages for comparing feedback. Early coverage focuses on selected consumer categories, and the summaries still depend on the authenticity of Reddit discussions.
• Spotify's AI DJ: Premium listeners in supported markets are getting nine new voice options and a setting that removes most of the spoken commentary between music selections.
• Apple's Claris Harbor: Claris is testing an early-access Mac app that runs downloadable AI models on Apple silicon hardware. It can connect compatible Macs over Thunderbolt 5 to share memory for larger models.
• Alsona: A new sales tool monitors more than 32 types of buying signals, ranks prospective customers against a business's target profile, and can draft or send outreach through AI agents. Users can review messages before they go out.

🏢 Big Tech & Major Companies

• ChatGPT voice is changing in December. OpenAI says Advanced and Standard Voice will be retired on December 11, 2026, with users moved to Live Voice. During the transition, advanced video and screen-sharing capabilities continue on the older voice mode until the new one supports them.
• Claude users received a one-time usage reset. Anthropic's developer account said paid Pro, Max, Ultra, and Team users can claim a free reset across models and usage limits before December 10 at 2 p.m. PT, after fixes for issues that consumed more usage than intended.
• NVIDIA's CEO faces continuing security scrutiny. The Edge Malaysia reported that Jensen Huang's acknowledged affair has prompted questions about a possible foreign intelligence operation. Allegations that a person involved was a spy remain disputed; NVIDIA's board is investigating, and no compromise of corporate secrets has been established publicly.

💼 AI Productivity, Labor & Economics

• Product managers are holding up better than designers in tech hiring. Lenny's late-2026 analysis found that product management openings increased faster than several other core tech roles, engineering hiring slowed, and design openings declined. It also found growing demand for AI-related skills and fewer remote vacancies.

🏛️ AI Policy, Governance & Safety

• The US and China plan a direct AI incident-alert channel. Treasury Secretary Scott Bessent described forthcoming talks on how to evaluate advanced AI models and communicate about serious incidents, even as the countries continue competing over the technology.
• OpenAI flagged a mismatch between student AI use and teacher guidance. OpenAI cited ThinkYoung research in which 79% of surveyed students reported using AI for assignments, but only 47% said they had received guidance about it. The company discussed educator training and ongoing research into learning outcomes.

📊 Fundraising & Deals Roundup

• Oura agreed to buy Counsel Health for $625 million. The deal combines Oura's wearable health information with an AI-assisted medical service that connects members with physicians. Bloomberg also reported a 153% fourth-quarter sales increase.
• OpenAI acquired Odessia, an AI travel-agent startup founded by former Sonder CEO Francis Davidson. Davidson confirmed he joined OpenAI, but the company had not detailed its travel plans and did not disclose acquisition terms.

Previous Around the Horn Digests

Catch up on everything you missed:

• Tuesday, November 10, 2026: Claude Fable and Mythos arrived; Anthropic's IPO reporting was corrected; Fireworks disclosed a breach.
• Monday, November 9, 2026: Anthropic's $3.4 billion power orders, Trump's comments about AI safeguards, and software-agent limitations.
• Friday, November 6, 2026: Faster AI research, new cybersecurity tools, and Claude's bank-account connections.
• Thursday, November 5, 2026: OpenAI's Linux vulnerability search, Claude's cell research, and Meta's coding acquisition.
• Wednesday, November 4, 2026: Security warnings at OpenAI, Google's school AI policy, and AI-designed hardware.
• Tuesday, November 3, 2026: Anthropic's research-acceleration warning, deceptive deepfake apps, and election safeguards.
• Monday, November 2, 2026: ChatGPT and Codex usage changes, advertising growth at Meta, and Apple's China plans.

That's a Wrap

If you reached the bottom, congratulations: you survived an IPO prospectus, a security proof with millions of lines, and an entire buffet of product launches.

For the shorter daily email, subscribe to The Neuron. See you tomorrow.

P.S. Know someone who'd enjoy this? Forward them the roundup and invite them to subscribe.