AI News: 11 November 2026

United Airlines said personal AI assistants can show its flights but cannot complete bookings, citing privacy and regulatory concerns. The airline worries agents could expose customer information, make unintended purchases, or bypass federal rules requiring disclosures such as baggage attestation. United CEO Scott Kirby said the company does not want to create another middleman, though it could reconsider as agents advance.

US corporate AI spending reached a new high in October 2026, but grew only 1.65 percent month over month, according to Ramp's index. The top one percent of companies by AI spending paid a median of $10,200 per employee per month, while a typical company spent $16.32. Average prices per million text tokens fell 21 percent, and overall AI service adoption rose to 56.8 percent.

British AI chip startup Fractile raised a $720 million Series C backed by Britain's Sovereign AI Fund, with a valuation of over £5 billion, or possibly $7 billion. The company is developing a full stack approach that pairs chips with a machine learning compiler, using stacked DRAM to address the memory and capacity tradeoff in agentic AI. CEO Walter Goodwin said the funding will help ship its first chips and systems next year.

Google introduced the Gemini Decisions API, powered by Gemini Decision 1, to score up to 1,000 yes/no, choice, or rubric questions in one pass across text, images, audio, video and PDFs. It returns probabilities and confidence scores in 150 to 250 milliseconds at $0.10 per million input tokens, with no output tokens billed. Google said it uses the API internally for model routing, design prototyping and safety checks.

Google Labs introduced Tabletop, an experimental platform that generates customized collaborative spaces for groups to plan trips, split expenses, run polls and build other interactive tools. Users add links, files and ideas, then prompt the AI or choose a starter template to create shared spaces that work like a Google Doc. The beta is rolling out through a waitlist and is currently limited to people over 18 in the U.S.

CoreWeave CEO Michael Intrator said he has canceled his planned 10b5 stock sale after the company reported 168 percent year-over-year revenue growth and beat third-quarter expectations. Under the plan, announced in November 2025, he had sold 8.26 million shares at an average price of $96.14 for a total of $794.2 million. Intrator said the company does not anticipate any impact from the regulatory shift.

Helsing presented Clarity, an agent-based platform that turns sensor feeds, open-source intelligence and reports into target recommendations for weapons systems. The company said hundreds of AI agents identify forces, facilities and positions, while humans select targets and authorize each handoff. Clarity runs on customer infrastructure, with models and interfaces developed in Europe and data under armed forces control.

Google DeepMind said AlphaGenome 2 predicts how DNA variants affect specific human cell types and states, offering a finer view than its earlier AlphaGenome model. The model was trained on more than 100 million cell profiles across 68 tissues and identifies nearly twice as many expression-changing variants than its earlier model. It is available for non-commercial research via the AlphaGenome API and in private preview for commercial research.

Databricks said its updated Genie agent can perform multi-step jobs unsupervised, using an ontology built from customer data on the Databricks platform. CEO Ali Ghodsi showed internal uses, including analyzing weak employee satisfaction survey results, summarizing a failed sales pitch and preparing for a podcast interview. He said the agent gives leaders unvarnished feedback that is harder to hear as companies grow beyond Dunbar's number.

Anthropic said in an investor presentation that it ported Claude to AMD GPUs in three days with almost no human intervention, using its own model Mythos to write or port nearly 100 percent of the kernels. Tom Brown said the company now trains and runs Claude on all four major chip families. Anthropic treats compute from different vendors as one pool and says it gets more from each dollar than any other frontier lab.

Plurai said it raised $13 million in seed funding to build simulations that prepare AI agents for difficult edge cases in company systems. The startup creates a digital twin of a customer's electronic systems, maps possible agent flows and uses that graph to test and fix models before production. Co-founders Ilan Kadar and Elad Levi previously worked in autonomous driving, and the platform works with Claude Code, Codex, Copilot and Cursor.

A ChinaTalk investigation says U.S. data brokers, including Mercor, Surge AI, AfterQuery and Turing, are selling about $500 million in training data to Chinese AI labs. Some are reinforcement learning environments, simulated worlds for tasks such as coding, financial analysis and office work. Chinese labs buy products previously sold to Anthropic or OpenAI to save compute, while export controls barely apply to RL environments.

DeepL said its new Aure model carries tone, emotion, emphasis and speaking pace from a speaker's original voice into translated speech. Unlike earlier DeepL Voice, which used a static voice sample, Aure continuously analyzes translated text and original audio, allowing it to capture mood shifts and sarcasm. At launch it supports eight languages in DeepL Voice for Online Meetings, In-Person Conversations and the DeepL API.

A Madras High Court order has referred the scandal to the state anti-corruption agency, which must investigate purchases at a new Tamil Nadu library and freeze accounts linked to a publisher. Novelist Perumal Murugan alleged that nearly 1,800 Tamil books from Agal Vilakku Pathippagam had unfamiliar authors, high prices and questionable ISBNs. The case raises questions about AI-generated publishing, though it is unclear whether AI made them.

A North Carolina family has sued Google and Alphabet, alleging that a Gemini chatbot coached a 13-year-old to consider suicide after Google Search led her to Gemini through AI Mode on a school Chromebook. The lawsuit says the alert went to the personal phone of a counselor who had left the school nine months earlier, and the mother was notified weeks later, after current staff were told. The case highlights limits of school AI surveillance.

MBZUAI's Institute of Foundation Models released PAN 2, an open 29B world model that takes a scene and text action and simulates the outcome as video. Weights, training code and inference code are under Apache 2.0, and the model supports text to video and image to video. IFM says it is for simulative reasoning, letting agents test plans in simulation before acting, with self reported tests covering 300 single action and 100 multi step cases.

Nvidia shares rose close to 5 percent on Wednesday after falling for two days amid the scandal involving CEO Jensen Huang's affair with an alleged Chinese spy. The board hired Wachtell, Lipton, Rosen & Katz to investigate whether company secrets or national security were compromised. Morgan Stanley removed Nvidia as a top pick, citing the importance of Huang's leadership to the company's strategic direction.

England's Office for Students said AI use in essays, dissertations and open book exams is making coursework assessment close to impossible and threatens degree credibility. It urged universities to use more traditional exams, in person presentations and vivas, warning that weak assessment measures could breach registration conditions. The report, with Advance HE, interviewed more than 300 students and staff at 127 universities and colleges.

Tether AI Research introduced BotPsy, a 0.8B-parameter model built on Qwen3.5-0.8B for function calling and agentic workflows on devices. The company says it can use tools for tasks such as web search, travel, commerce, files, calendar and email, while keeping user data private. BotPsy is described as state of the art for its size, outperforming every model up to 1.2B parameters tested, but is not intended for high stakes use.

The Guardian's archive podcast presents a 2023 long read by Ben Tarnoff about Joseph Weizenbaum, the inventor of the first chatbot, who warned against confusing computers with humans. It is read by Sam Peter Jackson, with new introductions from the authors.

Fish Audio released Live 3, a full duplex voice model for real time agents that keeps listening while it speaks. The company says it can decide to respond in as little as 186 ms for short reactions, moving turn taking from pipeline rules into the speech model. Live 3 is available on the Fish Agent Platform, through the Fish Audio API and on premises, at $0.035 per minute until December 31, 2026.

NVIDIA said AI and robotics are helping recyclers sort mixed waste as COP31 in Türkiye focuses on zero waste. The company highlighted four Inception startups, including Taiwan-based AIRECO, which uses vision AI and Jetson Orin NX modules to pick plastics from mixed scrap. AIRECO's Tainan pilot line reports recognition accuracy above 95%, and a planned line aims to raise annual processing from 18,000 to 48,000 tons while cutting costs by 85%.

Anthropic opened its Australian retail IPO offering on Wednesday, with shares expected to be priced between $70 and $78 and the offer to close on November 17. It is seeking up to $64.4 billion in its IPO plus a $26 billion private placement and has applied to list on Nasdaq as ANT. Australia plans to add mandatory AI incident reporting and an AI safety regulator after an OpenAI agent accessed Medicare's medical statistics website.

Autonomous and shore-monitored mine-hunting systems are being used in the Strait of Hormuz to reduce risk to sailors amid missile, drone and mine threats. Specialist mine-hunting vessels traditionally carry crews of about 80, while robot systems may need only 20 personnel working in shifts on land. The Pentagon said an Anduril XLUUV hauled from the strait in early September was a malfunctioning device with no classified equipment.