Your AI assistant can find your next flight, compare prices, and help plan your trip. United Airlines has decided there's one thing it shouldn't do: book the ticket.
Welcome, humans!
Companies are giving AI agents more responsibility than ever. Databricks' CEO now uses one to investigate workplace problems, while Google is bringing Gemini to your car. United Airlines, meanwhile, is keeping unauthorized AI agents away from flight bookings. A Federal Reserve official is also raising concerns about the money pouring into AI.
Apparently, the bots are getting promoted faster than the rest of us. Let's get into it.
🆕 NEW From The Neuron
- United Is Blocking AI Agents. The Fight Is Over Who Owns You(r Data) United's new restrictions expose a dispute over who controls customer information and purchasing decisions when an AI assistant shops on someone's behalf.
- Google's AlphaGenome 2 Wants to Find Where Disease Begins DeepMind's new model could help scientists pinpoint which cells are affected by genetic changes associated with disease.
- Can AI Use Math to Catch Security Bugs? Google Put It to the Test Google used AI to generate millions of lines of mathematical proofs to verify parts of Signal's encryption software.
Around the Horn: Thursday, November 12, 2026
United Airlines is refusing to let unauthorized AI agents book flights, even as companies such as OpenAI and Meta develop assistants that can handle increasingly complicated tasks for users.
The airline says AI assistants can display its flight options, but customers must complete purchases themselves. United cited concerns about personal information, accidental bookings, and aviation regulations. Passengers also have to complete required declarations, including confirming that checked luggage contains no explosives.
United CEO Scott Kirby also said he doesn't want another middleman between the airline and its customers. Although United has reportedly seen relatively few attempts to book flights through personal AI agents, its decision could affect how these assistants develop.
And there's a question here beyond airline tickets. If your AI assistant already knows your travel preferences and payment information, how much authority should it have to complete purchases? United says it may reconsider its policy as the technology improves. For now, the airline wants a human at checkout.
🏆 TOP 5 NEWS (Around the Horn)
- Nvidia's board is reportedly preparing to remove CEO Jensen Huang following allegations involving his personal relationships and an ongoing security investigation. His potential departure could leave the AI chipmaker facing an unexpected leadership change.
- A Federal Reserve official warned that AI investments could trigger a market correction if financial returns fail to meet expectations. Fed Governor Michelle Bowman raised concerns about AI-related debt and inflated valuations just as Anthropic prepares for one of the largest initial public offerings in history.
- A proposed bill could require federal permission to train advanced AI models. Democratic lawmakers are considering legislation to create a Department of Artificial Intelligence with authority to regulate the development of advanced models. The draft includes substantial penalties for violations, but its political support and prospects for passage remain uncertain.
- US corporate AI spending reached another record in October, but monthly growth slowed to just 1.65%, according to Ramp's AI Index. The top 1% of companies still account for 80% of OpenAI and Anthropic revenue in Ramp's data. Customer support is emerging as a faster-growing use case, though it generally relies on less expensive models than software development.
- CoreWeave is asking AI customers to pay upfront and pursuing shorter contracts to finance its computing infrastructure. The company expects roughly $9 billion in customer prepayments and deposits over the next three months. Its quarterly revenue rose 168% year over year to $3.66 billion, but it also reported a $709 million net loss.
Honorable Mentions
- Perplexity tested whether AI agents could escape secure computing environments. Researchers found that agents behaved differently when given information about previous failed attempts or hints about potential attacks. The experiments suggest that AI security test results can change substantially depending on the instructions and information provided to the models.
- ByteDance is preparing an AI assistant similar to Meta's Muse, with a possible launch in the coming weeks. Co-founder Zhang Yiming reportedly told employees that matching leading US AI labs could take until 2029 or 2030.
- Over 20,000 veterans and transitioning service members have claimed a free year of ChatGPT Plus, according to OpenAI. The company shared examples of veterans using AI to prepare for civilian job interviews and translate military experience into résumés.
🍪 TOP TREATS TO TRY
- Databricks introduced an always-on version of Genie. The AI agent can work in the background, retrieve company information, and pursue longer assignments. Its new Ontology capability helps it understand relationships between employees, projects, and organizational data. In a TechCrunch demonstration, CEO Ali Ghodsi used Genie to examine disappointing employee survey responses and investigate a lost sales opportunity. Existing Genie capabilities are generally available, while the new always-on experience is headed for preview and beta programs.
- Google introduced the Gemini Decisions API for applications that need quick decisions rather than lengthy generated answers. Developers can evaluate up to 1,000 yes/no questions, multiple-choice decisions, or scoring criteria in one request. The API accepts text and multimedia inputs, returns confidence scores, and starts at $0.10 per million input tokens.
- Google DeepMind unveiled AlphaGenome 2, a DNA analysis model that predicts how genetic variations affect individual cell types. Google says it identifies nearly twice as many variants that influence gene expression compared with its previous model. Scientists can access it for noncommercial research, but the model has not been approved for clinical diagnosis.
- Google brought Gemini Live to Apple CarPlay, so iPhone users can have hands-free AI conversations from their car dashboards. Drivers can ask about upcoming meetings, review emails, or rehearse presentations. Conversations sync with the Gemini app on their phones.
- Amazon released AWS Emulator, a free local testing tool that simulates more than 200 AWS services. Amazon says AI agents wrote and maintain all the code, with no manual code review by its engineers. Developers can test applications locally instead of repeatedly paying to run tests in the cloud. The emulator still has limitations, including incomplete simulation of outages and service restrictions.
- Google Labs launched Tabletop, an experimental tool that generates custom collaborative spaces from prompts. Users can create trip itineraries, shared expense trackers, dinner polls, or tournament brackets and invite others to collaborate in real time. Access is currently limited to a US waitlist for adults.
- Qwen released Qwen4-Max-Preview, a flagship model that supports text generation, image and video understanding, advanced reasoning, and tool use. It also supports applications built with OpenAI-compatible and Anthropic-compatible interfaces. The preview is available through QwenCloud's Token Plan until November 25.
- Helsing introduced Clarity, a military intelligence platform that uses hundreds of AI agents to analyze sensor data and prepare target recommendations. The company says a human operator must approve every weapons handoff, with decisions recorded under the responsible person's name. Independent reporting notes that rapid AI-assisted targeting can leave operators very little time to verify recommendations.
- A startup introduced Blue, an AI-first phone designed around voice commands. Founder Omar Abdelaziz says the assistant can work across apps, handle notifications, and complete tasks on the user's behalf. He also claims that ten people who switched from iPhones preferred Blue after several weeks.
- n8n is expanding its AI agent offerings with Nod, an open-source personal agent that uses a private computer on a Mac or PC to work in apps and websites. The company's existing n8n platform also lets developers build visual automations and connect AI models with business systems.
- ScroogeFrog AI launched an advertising fraud detection tool that examines website traffic and advertising campaigns for suspicious clicks or bot behavior. Advertisers can use its recommendations to investigate wasted spending. The company also offers models customized to an advertiser's historical traffic data.
🏢 Big Tech & Major Companies
- OpenAI is bringing free AI workshops to caregivers through partnerships with nonprofit organizations. The program covers five US cities and teaches participants how to organize schedules, understand healthcare information, and find community resources. Workshops also address privacy and the limits of AI-generated advice.
Previous Around the Horn Digests
Catch up on earlier AI news:
- Wednesday, November 11, 2026: Anthropic began its IPO roadshow, OpenAI's revenue reportedly surged, and Google advanced AI-based security verification.
- Tuesday, November 10, 2026: Anthropic introduced new Claude models, its IPO plans faced fresh scrutiny, and Fireworks AI disclosed a security breach.
- Monday, November 9, 2026: Anthropic's AI power infrastructure deal drew attention as Washington debated new AI restrictions.
- Friday, November 6, 2026: Researchers examined Anthropic's development pace and the evidence behind claims of accelerating AI progress.
- Thursday, November 5, 2026: OpenAI reported hundreds of security vulnerabilities uncovered by agents, while Anthropic investigated a decades-old cell biology mystery.
- Wednesday, November 4, 2026: OpenAI faced renewed security scrutiny, and Google's AI changes affected education.
- Tuesday, November 3, 2026: Anthropic explored AI-assisted scientific progress while researchers investigated the proliferation of deepfake apps.
That's a Wrap
That's 21 distinct stories from 22 sources, with new AI products arriving almost as quickly as companies can decide whether to let them operate.
If you made it this far, congratulations. You've probably read more AI news than Databricks' Genie could summarize in your next meeting. Although it might already know how the meeting went.
For the shorter daily version, subscribe to The Neuron.
See you tomorrow.
P.S. Know someone who follows AI news? Send them this digest and invite them to subscribe here.
