AI-generated

OpenAI disbands its robotics team as hackers adopt agentic AI frameworks

Sunday, 15 November 2026 | 53 articles

OpenAI has shut its standalone robotics team and moved researchers into core groups, betting that general-purpose models will drive embodied AI. Meanwhile, the FBI warned that agentic AI hacking frameworks such as ARTEX are going mainstream, and LeCun's latest work is reshaping the AI scaling debate.

Listen to brief as podcast
Martin Ševčík

OpenAI has just dissolved its robotics team for the second time, and I find the reasoning more interesting than the retreat itself. According to reports, the cause wasn't a shortage of data, which was the usual excuse the first time around in 2020. This time the bet appears to be that general-purpose models like GPT-6 Astra will simply absorb the problem. The researchers have been folded into the groups under Mark Chen and Greg Brockman, and the specialized robotics effort is gone. I think this is a genuinely bold wager, and I'm not convinced it's right. Physical systems punish errors in ways text never does, and a model that writes a fine paragraph can still knock a glass off a table. But if OpenAI is correct, a lot of narrow embodied-AI startups just lost their moat.

It's the same philosophical fault line that runs through the scaling debate. A new discussion of Yann LeCun's work claims it changes the math on scaling and lends weight to the Platonic Representation Hypothesis, the idea that sufficiently large models converge on a shared picture of reality regardless of how they're trained. The source I saw is thin on detail, so I'd treat the framing with caution. Still, notice the tension: OpenAI is betting that one general model will do everything, while LeCun has long argued that scale alone isn't the path. Both can't be fully right, and robotics may end up being the test case. By the way, Multiverse Computing's experiment with multi-head quantum neural networks inside a 30-billion-parameter LLM shows how many people are looking for an architectural shortcut beyond brute-force scaling. It's early, and I'd wait for results before getting excited.

Meanwhile, the agent story is moving from promise to problem. The FBI issued a TLP:CLEAR FLASH report on November 13 about ARTEX, an autonomous hacking framework, and the key point is the trajectory: agentic attack tools are graduating from penetration-testing utilities to a mainstream threat. Defenders are responding in kind. CISOs now argue that manual oversight can't keep up with AI agents and that governance itself has to become autonomous. That sounds sensible until you ask the obvious question: who governs the governors?

Money is already flowing toward the answer. LangChain's reported $3 billion valuation shows investors shifting toward tools that let businesses manage, monitor and control autonomous agents. I read that as a signal that control, not capability, is becoming the scarce layer of the stack. For anyone building with agents, the practical lesson is that observability and permissions deserve as much design time as the prompts do.

So here is what I'm left wondering. If general models really can swallow robotics, and agents are both the next product wave and the next attack surface, does the real competitive advantage shift from who builds the smartest model to who can be trusted to keep it on a leash?

List of sourced links used in the brief

Ex-OpenAI board member who ousted Altman on AGI skepticism and their friendship

The former OpenAI board member who took part in firing Sam Altman says Altman is firmly convinced AGI is coming, while she is not. She describes the difference in outlook as a healthy dynamic that lets them remain friends. Source: theverge.com

Multiverse Computing trials quantum neural networks inside a 30B LLM

Multiverse Computing is testing multi-head Quantum Neural Networks integrated into a Large Language Model with 30 billion parameters. Source: quantumzeitgeist.com

Karpathy-inspired prompt rules aim to fix common LLM coding mistakes

A minimalist open project distills Andrej Karpathy's critiques of LLM coding pitfalls into a compact set of rules. The idea is that a single simple prompt can noticeably improve the quality of AI-generated code. Source: medium.com

Altman: running AI locally is a poor use of scarce memory

Sam Altman reportedly argues that running AI models on personal computers wastes scarce memory. Meanwhile, Anthropic is said to be preparing for a major IPO. Source: youtube.com

Repligate shares a post on local AI as a foundation to build on

A repost from the account Repligate nods to a discussion of the ontologies of the artificial. It also includes a line arguing that local AI is a dependency developers can reliably build on. Source: digg.com

Runway's open-source Praxis-1 turns video AI into a robot controller

Trained on video and robot data, the model takes in camera images, text instructions, and position data. From these inputs it calculates matching robot movements. Source: the-decoder.com

Anthropic: old prompting habits are holding Claude back

Anthropic has made Claude Opus 5.5 the default model on all paid plans. Alongside it, the company published a guide that retires outdated prompting habits. Source: the-decoder.com

XYBench tests whether AI can spot the flawed premise behind a question

XYBench is a new benchmark of more than 8,000 questions that checks how AI assistants handle requests built on misconceptions, such as asking how to thicken frosting. It measures whether a model addresses the underlying problem rather than just answering the question as posed. Source: huggingface.co

RunanywhereAI releases TrnRT, an open inference engine for Trainium LLMs

TrnRT is an open inference engine for running large language models on Trainium hardware. It bundles model execution, caching and an API server into a single system. Source: news.lavx.hu

South Central Railway upgrades Bholu AI assistant with an indigenous language model

South Central Railway has upgraded Bholu, its AI-powered passenger information assistant, with an indigenous language model. The Mini and Max versions were introduced in May at Kacheguda and Malkajgiri railway stations. Source: deccanchronicle.com

Prompting tweaks help AI seizure localization, but only for certain models

A study tested prompt debiasing and example retrieval across seven open-weight language models for seizure semiology localization. Performance gains turned out to be model-specific rather than universal. Source: bioengineer.org

Hackers feed stolen LLM API keys to ARTEX AI agents to keep stealing credentials

Chinese-nexus threat actors used ARTEX, an open-source agentic penetration-testing framework, to steal LLM API credentials. The stolen keys are then fed back into the tool to power further attacks. Source: gbhackers.com

Survey examines how LLM reasoning generalizes beyond training data

Large language models perform strongly on mathematical, logical and scientific reasoning tasks, but this progress is often measured within familiar data. A new survey reviews how well such reasoning generalizes outside the training distribution. Source: alphaxiv.org

KAIST boosts lip-reading AI accuracy using a diffusion language model

KAIST announced on the 15th that a team led by Professor Noh Yong-man of its School of Electrical Engineering has improved the accuracy of lip-reading AI with a diffusion language model. Source: biz.chosun.com

Claude's Duet agent is a bit neurotic, but surprisingly useful

Duet is Anthropic's consumer-grade AI agent. It works as a helpful partner for catching things that might otherwise slip through the cracks. Source: theverge.com

Emad Mostaque on Jensen Huang's departure and NVIDIA's succession

In a podcast discussion, Emad Mostaque weighs in on Jensen Huang's departure and what it could mean for NVIDIA. The conversation also explores broader implications for the future of AI. Source: youtube.com

Altman reveals the coolest part of leading OpenAI: holding a new AI computer

Sam Altman has shared what he considers the most exciting aspect of running the world's leading AI company. According to the coverage, it involves the thrill of holding a new AI computer in his hands. Source: youtube.com

More Large Language Models news

FBI Warns of ARTEX: Agentic AI Hacking Frameworks Go Mainstream

The FBI issued a TLP:CLEAR FLASH report (FLASH-20261113-001) on November 13, 2026, giving a threat overview of ARTEX, an autonomous hacking framework. The report signals that agentic AI attack tools are moving from penetration-testing utilities to a mainstream threat. Source: oodaloop.com

CISOs: Securing AI Agents Demands Autonomous Governance

As Generative AI adoption accelerates and reshapes how people interact with technology, security chiefs argue that AI agents can no longer be controlled by manual oversight alone. CISOs say governance itself needs to become autonomous to keep pace. Source: cybersecurity-insiders.com

LangChain's $3 Billion Valuation Spotlights AI Agent Control

LangChain's new valuation shows investor money shifting toward tools that let businesses manage, monitor and control autonomous agents. Control over agents is becoming a central piece of the AI stack. Source: wersm.com

Personal Agent Protocol 1.0 Released

Version 1.0 of Personal Agent Protocol, known as Poppy, is now live. It is enabled for everyone using Muse, Meta's AI agent, and for every merchant on the platform. Source: sierra.ai

AI Agent on a Local LLM: Measurements From a Single GPU

Running an agent on your own GPU sounds easy: point it at Ollama and stop paying per token. The author set up OpenClaw, an open-source AI assistant, with a 9B model and measured how it actually performs. Source: sitepoint.com

Before AI Does Your Holiday Shopping: 7 Security Rules

A security expert advises connecting a credit card rather than a debit card to a shopping AI agent, capping how much it can spend, and keeping purchase approval switched on. The tips aim to limit the damage if an agent goes wrong. Source: securityboulevard.com

Setting a Context Window Budget Before Long Agent Tasks Fail

Agents get slower and costlier as tool results accumulate in their context. The article lays out reserve, prune and summarize thresholds that help long-running LLM agents stay accurate. Source: startupfortune.com

More AI Agents & Automation news

Dwarkesh Patel: America's AI lead over China is hours, not months

Patel argues that recursive self-improvement in AI could trigger an intelligence explosion and yield billions of human-level AIs by 2030. He says the US labs' edge over China amounts to only months, or even less. Source: finance.biggo.com

Geoffrey Hinton warns of existential AI threats as superintelligence nears

Nobel laureate Geoffrey Hinton, often called the godfather of AI, cautions that the race toward superintelligence poses existential risks. In the conversation he also discusses Nvidia's Jensen Huang. Source: thirteen.org

Advocacy groups sue five AI firms to halt new model development

A coalition of online safety and consumer advocacy groups filed suit in New York on Thursday against five of the world's leading AI companies. The plaintiffs want development of new AI models stopped, citing major risks. Source: arkansasonline.com

Study finds sparse attention heads that encode goals inside LLMs

Northeastern University researchers report that a small set of sparse attention heads in large language models causally encodes the model's internal task goals. The preprint adds to interpretability work on how LLMs represent what they are trying to do. Source: casrai.org

How America could sabotage its own AI lead

A mix of political, commercial and ideological pressures threatens to erode America's lead in AI. The piece argues that public controversies are strong enough to derail the country's leadership in the field. Source: cset.georgetown.edu

Opinion: Voters took on data centers — now it's time to govern AI

Data centers were not formally on this year's ballots, yet candidates critical of such projects won by wide margins. The op-ed argues this voter signal should now push policymakers toward governing AI itself. Source: startribune.com

More AI Safety & Alignment news

MIT builds generative AI copilot for structural engineering design

MIT researchers created an AI copilot for structural design that pairs human guidance with topology optimization. The approach raised load capacity by 39 percent. Source: casrai.org

Guide to measuring Cursor vs GitHub Copilot market share in 12 steps

A 70-minute, 12-step audit compares the market share of Cursor and GitHub Copilot. It ships with a Python kit that combines survey, seat and pulse data and computes confidence intervals. Source: shattered.io

New AI copilot lets engineers steer structural designs as algorithms generate them

Algorithmic design has already produced parts optimized for ultralight aircraft, high-bandwidth antennas and F1 racing cars. A new AI copilot now lets engineers guide the process while the algorithms create the structure. Source: techxplore.com

Cordial's headless CDP brings customer data queries into ChatGPT, Claude and Copilot

Cordial's headless CDP lets marketers explore and activate customer data directly from Claude, ChatGPT and Copilot. The article also outlines what CDP buyers should check next. Source: marketscale.com

More AI Tools & Products news

SpaceX AI Unit Cuts Around 60 Jobs at Grok Imagine

SpaceX's AI division has laid off roughly 60 employees from its Grok Imagine video and image generation team. Only a small group remains to keep the product running. Source: finance.biggo.com

More Image & Video Generation news

OpenAI shuts its standalone robotics team and folds researchers into core groups

On November 13, OpenAI disbanded its standalone robotics team and placed the staff under Mark Chen and Greg Brockman. It is the company's second retreat from robotics since 2020. Source: startupfortune.com

OpenAI moves robotics researchers to other teams, betting general models will suffice

OpenAI is dissolving its standalone robotics team for the second time, reportedly not due to a lack of data. The company appears to believe general-purpose models such as GPT-6 Astra may make specialized robotics AI unnecessary. Source: the-decoder.com

OpenAI drops its robotics team as general-purpose models take the lead in embodied AI

OpenAI confirmed on November 13 that it dissolved its standalone robotics research team. Its staff were folded into the company's core research and applications groups. Source: finance.biggo.com

Embodied AI robots take over night-shift parcel sorting in Shenzhen

Three embodied AI robots joined a night-shift parcel-sorting line in Shenzhen, in China's Guangdong Province. They were deployed during the November online shopping rush. Source: news.cgtn.com

Robust AI's Carter robot is designed to work alongside people

MIT News (13 Nov 2026) looks at how Robust AI's Carter cart robot is built around human-robot interaction research. The piece also considers what this means for research offices. Source: casrai.org

More Robotics & Embodied AI news

LeCun's latest work reshapes the AI scaling debate

A new discussion of Yann LeCun's work argues it changes the math on AI scaling and lends support to the Platonic Representation Hypothesis. The snippet offers few details beyond that framing. Source: youtube.com

Hassabis: schools must be rebuilt for the AGI era

Demis Hassabis says education in the age of AGI should center on people skills, projects, creativity and judgement. Rote learning, he suggests, can happen outside the classroom. Source: youtube.com

Qubit States Mapped to Phase-Space Probability Distributions Using Normalizing Flows

Researchers have used Normalizing Flows to map qubit states onto probability distributions in phase space. The approach offers a new framework for classically simulating quantum systems. Source: quantumzeitgeist.com

Germany Backs QUDORA's Quantum Machine Learning Project for Energy

Germany's Federal Ministry of Research, Technology and Space (BMFTR) is funding QUDORA's new project, QUANDA. The project will integrate the company's trapped-ion technology with machine learning for energy applications. Source: quantumzeitgeist.com

Datasets powering AI weather models are now open to researchers

The team is publishing two datasets, PARDIG and SatWeave, that underpin its AI weather models. The release is meant to give researchers and developers a foundation for building new weather, hydrology, and climate applications. Source: deepmind.google

Data Type Decides Whether Classic ML or Deep Learning Wins at Breast Cancer Prediction

A two-track comparison finds that classical machine learning delivers better accuracy on structured clinical biopsy tables. Deep learning, by contrast, appears to gain its edge on other types of data. Source: bioengineer.org

PhishNet++ Brings Explainable Multi-Branch Deep Learning to Phishing URL Detection

Phishing remains one of the most common cyber threats, causing major financial losses and exposing sensitive user data. PhishNet++ is a multi-branch deep learning framework designed to detect malicious URLs robustly while explaining its decisions. Source: link.springer.com

More AI Research news

Mercor raises $750M Series D at $20B valuation

Mercor announced a $750 million Series D round on 10 November 2026, valuing the company at $20 billion. Its marketplace pays domain experts for their work, and the new funding underscores investor appetite for expert-driven AI talent platforms. Source: casrai.org

More AI Business & Funding news

Chinese AI labs reportedly train models on restricted Nvidia Blackwell chips

US export controls aren't keeping Chinese AI labs away from restricted Nvidia hardware, according to a report from a former OpenAI engineer. Will DePue is the source of the claim, which suggests the chips are still reaching labs in China despite the restrictions. Source: theregister.com

Musk: Nvidia will lose 'critical' deals without Jensen Huang

With Huang gone, markets are asking whether Nvidia will keep bankrolling the AI boom. Elon Musk says the company stands to lose key deals without him. Source: reuters.com

Nvidia CEO Jensen Huang resigns as Congress probes AI chip dealings

Jensen Huang has stepped down as Nvidia CEO and agreed to testify before Congress. The move follows his alleged affair with a suspected Chinese spy. Source: fox10phoenix.com

Three AI chip stocks beyond Nvidia that retail investors may be overlooking

Jensen Huang's abrupt departure has shaken the AI hardware narrative and shifted attention to the quieter companies that actually build and supply the chips. Source: finance.yahoo.com

The fall of the man behind Nvidia's $5.7 trillion AI empire

The founder and CEO made Nvidia the linchpin of the AI boom. The open question now is whether the company can hold together without him. Source: fortune.com

More Hardware & Infrastructure news

Support the project

AIskimIQ is an independent project. If you find it useful, you can support its development with a coffee.

Buy me a coffee ☕