This issue collects 36 AI updates from 2026-11-13, led by AMD Unveils New AI Data Center GPU, Shares Jump More Than 5%, After Huang's resignation, Trump floats Tim Cook to run Nvidia, Northern Trust and Microsoft deploy an AI agent that cuts onboarding prep from weeks to days, Bielik 11B v3.1 ships as a 56 language Polish open model.
AMD introduced a new AI data center GPU and its stock rose more than 5% in a single session, according to a TipRanks report.The public summary ties the product launch to the share move but gives no model name, specifications or shipping timeline.
SoftBank lost more than $3 billion pre ordering natural gas turbines intended to power AI data centers, TechCrunch reported.The loss raises questions about how far a gas fired buildout can carry the AI boom and what it means for power markets.
Nebius laid out roughly $40 billion in planned 2027 spending on its third quarter earnings call, far above Wall Street expectations, to be funded largely through customer prepayments and loans backed by customer contracts.The plan is already being tested by months of delays at a Microsoft leased New Jersey data center, and the stock fell 17% despite revenue growing 571% year over year.
President Trump posted on Truth Social that only one man could replace Jensen Huang as Nvidia's chief executive: Apple's Tim Cook, writing that “A Legend replaces a Legend!” The suggestion follows Huang's resignation on Thursday and a Time op ed by Yale's Jeffrey Sonnenfeld recommending Cook, though CEO succession at Nvidia is a board decision rather than a presidential one.
Phota Labs has released an iOS app that uses AI to turn everyday snapshots into professional looking photographs while keeping people's real appearance intact.The a16z backed startup, founded by two former Adobe computational photography engineers, pitches it as a photographer that works before you ask.
On November 12, Google said Gemini 4 Hydrogen is now live in Search for users worldwide, describing it as the biggest model upgrade to Search yet.The update deepens intent understanding, query fan out and structured answers that include images, comparison tables and generated tools.
Reports on November 12 said Google's Gemini 4 Hydrogen outscored the larger Argon just four weeks after launch, while another report described Hydrogen as a fast, efficient model under internal testing.The specific benchmarks and the size of the gap were not disclosed.
Philippine carrier Globe has announced AI Fiesta, a unified platform bundling models from Claude, Gemini and ByteDance with tools for images, video, presentations and websites, rolling out from November 15.It adds automatic model routing and a multi model consensus tool, letting subscribers bypass the paywalls that normally guard premium AI services.
Chinese medical AI company Future Doctor has released results from a large evaluation of its MedGPT model on complex lung cancer cases, with its strategies preferred over those of 250 hospital physicians by 88 to 12.In the same condition comparison, MedGPT scored 96.22 against GPT 5.
Northern Trust has expanded its Microsoft collaboration with an AI agent that prepares clients' historical investment and transaction data for onboarding on its Front Office Solutions platform, cutting data preparation time by 83%, from about two and a half weeks to three days.The agent runs on Azure and Microsoft Foundry, integrates with Excel and SharePoint, and reports a 96.5% first pass F1 quality score, with analysts reviewing its output.
Trust Stamp has launched Knowing Consent, a biometric approval layer that verifies a bank customer in person before a transaction initiated by an AI agent is authorized.The system separates an agent's access to an account from the account holder's approval of one specific payment.
Crypto Briefing reported on November 12 that Meta's personal AI agent Muse can now make phone calls on behalf of users in the United States.The step moves the assistant from answering questions to executing real world tasks that require live conversation.
OpenAI published a customer story on Amplitude showing how its proactive product agent Wave, roughly 80% powered by OpenAI models, turns analytics and session replays into shipped product changes.Amplitude says Wave cut improvement cycles from about a month to one or two days, and a personalisation experiment on its own homepage lifted card driven messages 160% in a day.
OpenAI launched Daybreak Red Team and Daybreak Defense Factory on November 12, pitching agent fleets that pen test web applications and generate merge ready patches for human review.President Greg Brockman argued that security now depends on which frontier model a company defends with, on the same day nonprofits sued OpenAI and rivals to halt more capable systems.
Scale AI's Remote Labor Index, updated November 12, ranks Anthropic's Fable 5.2 first with 41.3% of complete freelance projects judged to meet a professional standard, nearly double Opus 5.
Commvault introduced an AI agent recovery plan aimed at IT and security teams, designed to restore AI agents and roll back unwanted changes those agents trigger across environments.The announcement is short on product specifics, but it signals that agent state and agent caused changes are now being treated as recoverable production assets.
Feedzai has launched a capability called AI Agent Detection aimed at automated financial crime, including attacks that get around multi factor authentication, FF News reports.The launch reframes fraud defence around agent behaviour instead of the older split between humans and scripts.
openJiuwen has released an RSI native AgentOS, a runtime that opens extension points across the interface, gateway, server and agent execution layers so agents can build and install new capabilities on themselves.Its WorkSwarm app loads agent generated extension packages without changing existing code, with version management and rollback when a new build breaks something.
Poland's Bielik 11B v3.1 is out with support for 56 languages, widening the multilingual footprint of Europe's open weight model family.The instruction tuned release ships under the Apache 2.
Hugging Face's chief executive has publicly pledged to keep fighting for open source AI after Jensen Huang's exit, according to TechCrunch.The remarks land amid questions about Nvidia's $12.9 billion acquisition, putting the governance of the open model ecosystem back in the spotlight.
IBM's Granite team released granite 4.1 3b LLMON preview and granite 4.1 8b LLMON preview, two experimental models trained to read and write a markup language called LLMON that separates instructions from data explicitly.
The llama.cpp project published a Hugging Face blog post introducing version 1.0.Watch whether this signal spreads into products, developer ecosystems, or enterprise buying decisions.
QbitAI reports that 无界前沿 has open sourced Fx Bio 1031, a life sciences model it says outperforms GPT 6 Astra and Opus 5 on difficult blind tests.The release emphasises data staying on premises and local deployment, with RSI framed as the guiding idea.
UK AI chip company Fractile has raised $720 million, according to an eWeek report, framing the round as a direct challenge to Nvidia's AI chips.The size of the raise stands out for a British semiconductor startup and adds a well capitalised contender to the fight over AI compute.
The Wall Street Journal reports that some small startups have handed the chief marketing officer role to AI agents rather than treating them as writing assistants.The shift pushes agents into executive decision making and raises hard questions about accountability, brand judgment and human oversight.
Nectar has launched an AI agent that helps brands find creators, run outreach and manage partnerships, then track how campaigns performed.The product packages the most manual parts of influencer marketing into a single automated workflow.
IBM has acquired AuthMind to bring identity observability and threat detection into HashiCorp Vault, extending protection across the full lifecycle of AI agents.The companies had worked together since IBM Ventures joined AuthMind's 2023 seed round; financial terms were not disclosed.
NFON AG released AI Agent Studio on Thursday, letting companies build custom AI agents either by describing tasks in natural language or by assembling visual workflows.The product is available now and is sold through NFON's European partner network to small and mid sized businesses.
United Airlines said it will not allow customers to use AI agents to book flights or manage reservations, citing cybersecurity risk.The decision has a major carrier explicitly rejecting browser operating agents in one of the highest frequency transaction scenarios.
Nvidia CEO Jensen Huang resigned on November 12 after admitting to an affair with a woman described as an alleged Chinese spy, raising U.S.national security concerns.
Taiwan’s Non Store Retailers Association has publicly called for three rules governing AI agents that place orders, according to BigGo Finance, and the Ministry of Digital Affairs has agreed to study a risk framework.The dispute centres on authorisation, liability and platform rules once software buys on a shopper’s behalf.
Nvidia representatives told Rep.Bill Foster that the chip tracking mandates in the Chip Security Act are technically achievable, according to an interview with MIT Technology Review, though nothing was ever put in writing.The private assessment sits awkwardly beside the public position of the Semiconductor Industry Association, of which Nvidia is a member, which called the requirements “new, untested, and potentially infeasible.
A coalition of nonprofit groups has sued OpenAI, Google, Anthropic and other frontier developers in San Francisco Superior Court, seeking an injunction that would halt work on models more capable than today's systems.The complaint relies on California's unfair competition law and a public nuisance theory, arguing courts can intervene before a mass casualty incident rather than after one.
Democrats on the House Ways and Means and Senate Finance committees opened an inquiry Thursday into how the Trump administration handled Nvidia's chip sales to China.In a letter to Commerce Secretary Howard Lutnick, they described an arrangement that routed Taiwan made H200 chips through the US before export and asked why Megaspeed kept buying after intelligence warnings.
A November 12 analysis by Unemployment Labs says weight initialization patterns in six open weight models from DeepSeek, Moonshot AI, MiniMax and Ant Group match Nvidia's export controlled Blackwell GPUs.The report stresses that the evidence covers initialization only and cannot prove which chips trained the models, who owns them, or how they were obtained.
Fujian's provincial government office issued an AI action plan for 2026 2028 on November 12, calling for 50 AI products, more than 100 application scenarios and at least 60 industry specific vertical models.The plan also maps city level data centers around Fuzhou and the Xiamen Zhangzhou Quanzhou corridor into a province wide computing network.
