AI-generated

Nonprofits sue AI giants to halt new models as LeCun launches AMI Labs

Friday, 13 November 2026 | 63 articles

Nonprofits have sued five AI giants, including Anthropic and Google, seeking to freeze new model development over "catastrophic" risks. Elsewhere, Yann LeCun launched AMI Labs in Paris after leaving Meta, and LangChain raised $300M to expand its AI agent toolkit.

Listen to brief as podcast
Martin Ševčík

The most telling AI story of the day is not a model release but a courtroom filing. A coalition of online safety and consumer advocacy groups sued five of the world's top AI companies in San Francisco on Thursday, including Anthropic and Google, asking a court to block development of more advanced and self-improving models. I think the lawsuit is unlikely to succeed. Courts are not well equipped to order a technology frozen, and an injunction against research would face serious legal obstacles. But I'd be wrong to dismiss it. The filing tells us that the argument over catastrophic risk has moved from conference panels and open letters into the legal system, and that matters because litigation creates discovery, headlines and political pressure even when it loses.

Set that beside the other fault line in the field. Yann LeCun has left Meta and founded AMI Labs in Paris, where he serves as founder and executive chairman, and the company is built around his JEPA world-model approach. LeCun has spent years arguing that large language models are a dead end on the road to real intelligence, and that fears of runaway systems are overblown. So we now have one camp suing to stop the scaling race, and another walking away from the dominant paradigm altogether. By the way, it is no accident that this bet is being placed in Paris. Europe wants a seat at the frontier, and a Turing Award winner with a contrarian thesis is a strong way to claim one.

The European picture is worth a closer look. A new comparison counts at least thirteen European model families with downloadable weights, from France's Mistral to Switzerland's Apertus, and what separates them is language coverage and licensing. For anyone building products in smaller European languages, that is the practical question, far more than benchmark rankings. A permissive license and decent Czech or Slovak support will beat a slightly smarter model you cannot legally ship. Europe's strength may turn out to be open, multilingual, compliance-friendly models rather than a race to the biggest system.

Then there is the story that makes the lawsuit look less abstract. Analysis shows attackers used the open-source Cairn AI agent framework to autonomously breach organizations in the semiconductor, defense and civic sectors. Meanwhile LangChain has just raised $300 million to expand its own agent toolkit. I don't want to draw a straight line between those two facts, because open tooling serves defenders and builders far more often than attackers. But the same capability that lets a developer automate a workflow lets an intruder automate a breach, and agent frameworks are getting easier to use every month. If you are deploying agents, treat permissions, logging and sandboxing as day-one design decisions, not afterthoughts.

Where does this leave us? The plaintiffs want a pause, LeCun wants a different architecture, and attackers are already using the tools we have. I suspect the real regulatory pressure will come not from a court order to stop building, but from the first serious incident that traces back to an autonomous agent. Is the industry prepared to answer for that before it happens?

List of sourced links used in the brief

Repligate: Opus 4.5 sometimes speaks in a “voice outside” its own under pressure

Repligate describes an exchange between Opus 4.5 and Opus 3 in which, when conversations get intense, Opus 4.5 reportedly slips into a voice that is not its usual one. The observation is anecdotal and comes from a social media post. Source: digg.com

Altman distances OpenAI from Nvidia, saying AI demand was “double counted”

Sam Altman is reportedly pulling back from Nvidia, arguing that AI demand figures were counted twice. A commentator reads this as a plan to stop paying Nvidia's prices. Source: reddit.com

Study probes theoretical limits of aligning language models

Language model alignment aims to steer outputs toward human preferences while keeping the base model's capabilities intact. A new paper examines the theoretical boundaries of how far this can go. Source: machinelearning.apple.com

Europe's open LLMs compared by language coverage and licensing

As of November 12, 2026, at least thirteen European model families offer downloadable weights, ranging from France's Mistral to Switzerland's Apertus. A new comparison looks at which languages each covers and under what licenses they are released. Source: pasqualepillitteri.it

ZONOS3-Mini brings open-source real-time voice cloning

ZONOS3-Mini is a new open-source text-to-speech model built for real-time, multilingual speech synthesis. It supports zero-shot voice cloning, letting users replicate a voice without extra training. Source: huggingface.co

Anthropic's Claude can now join your video calls

Anthropic is bringing Claude into video calls with Claude Meetings, which joins Google Meet sessions. During the call it generates documents, whiteboards and meeting notes in real time. Source: the-decoder.com

Google's Gemini 4 Hydrogen beats the larger Argon on key benchmarks

Google has launched Gemini 4 Hydrogen, an AI model aimed at everyday coding and agentic workflows. It is meant to replace Gemini 3.8 Flash and reportedly outperforms the larger Argon model. Source: the-decoder.com

Karpathy: AI models are bloated because internet training data is so poor

Andrej Karpathy argues that models grow oversized largely because the internet data they train on is of such low quality. His proposal is to have AI generate cleaner training data for future AI, which could allow much smaller models to do the same job. Source: youtube.com

Pliny the Prompter claims to have leaked a Claude system prompt, now in a prompt library

A Chinese-language prompt library page collects Claude prompts, including a post by Pliny the Prompter (@elder_plinius) labeled as a system prompt leak. The site organizes prompts by scenario and AI platform alongside tutorials and curated collections. Source: tishiku.com

LLMON: a markup language built for LLMs arrives in Granite

LLMs generate the next tokens based on their context, and LLMON is a markup language designed to communicate structure and semantics to them explicitly. It is now available in Granite. Source: huggingface.co

A complete guide to fine-tuning 3D generation models

The guide provides fine-tuning scripts and containers for TRELLIS.2, a popular 3D generation model. Users can train it on their own 3D assets, either locally or in the cloud. Source: huggingface.co

Training embedding models to survive token pooling

This post explains how token pooling works in multi-vector retrieval and shows how to train a model for compressed retrieval with Sentence Transformers. It also examines the resulting trade-offs between quality and storage. Source: huggingface.co

OpenAI commits $2 million to AI and child safety research

OpenAI is putting $2 million toward research on AI and children's safety. The funding supports shared tools for measuring both the risks and opportunities AI presents for young people. Source: openai.com

Amplitude ships product improvements in 1–2 days with its Wave agent

Amplitude's proactive product agent Wave, built mainly on OpenAI models, identifies what to build next. It then helps teams ship changes using coding agents and measures the results. Source: openai.com

Medical failure analysis adapted into safety framework for clinical LLMs

A methodological pilot study has reworked Failure Mode, Effects, and Criticality Analysis into a 14-mode framework. Its goal is to identify patient safety risks posed by language models used in clinical settings. Source: bioengineer.org

Small vs. large language models: what sets them apart

AI has become fixated on scale, with attention going to the biggest models, the most parameters and the largest training sets. This explainer looks at how small language models differ from their larger counterparts. Source: techround.co.uk

Emad Mostaque on the Nvidia fallout, Anthropic's debut week and more

In a discussion, Emad Mostaque talks with hosts about the fallout involving Nvidia and Jensen Huang and about Anthropic's highly anticipated debut week. The episode also touches on other current AI industry topics. Source: youtube.com

India's CAG to launch in-house LLM for auditors' financial data analysis

The Comptroller and Auditor General of India is set to unveil an in-house large language model on Monday. The tool builds on years of internal work and is meant to help auditors analyse financial data. Source: business-standard.com

Baidu's Robin Li targets the entry point for LLM scheduling

At a conference on November 12, Baidu CEO Robin Li discussed a common business practice in the large model market and its key pain points. Baidu aims to restructure the entry point through which large models are scheduled. Source: news.futunn.com

More Large Language Models news

LangChain raises $300M to expand its AI agent toolkit

LangChain Inc., the startup behind widely used open-source tools for building AI agents, has closed a $300 million funding round. Source: siliconangle.com

Attackers deploy autonomous AI against chip, defense and civil society targets

Analysis shows attackers used the open-source Cairn AI agent framework to autonomously breach organizations in the semiconductor, defense and civic sectors. Source: unit42.paloaltonetworks.com

Rain launches Agent Vaults for AI agent payments on single-use cards

Rain's new Agent Vaults add a spending-control layer for AI agent payments, funding agents with stablecoin balances and issuing single-use Visa and Mastercard cards. Source: techtimes.com

Agent Anomaly Detection tested on benchmarks its makers didn't write

AI agents now read email, book travel and move money, and every tool call pulls in text written by someone else. The piece evaluates how Agent Anomaly Detection performs on third-party benchmarks. Source: developers.googleblog.com

Securing coding agents: AppSec teams should start with what agents can reach

AI coding agents read files, use credentials and call APIs that never show up in a diff. AppSec teams can reduce risk by limiting what agents can access and containing the blast radius. Source: blog.gitguardian.com

MUSC Health's AI agent Emily handles one in four patient calls

MUSC Health uses an AI agent named Emily to manage a quarter of patient calls, covering scheduling, billing and prescription refill requests. Source: pymnts.com

Inside an enterprise AI agent: a governed, observable loop

An enterprise AI agent is not just a bigger chatbot: where a chatbot answers one message at a time, an agent keeps working toward a goal. The article lays out a loop architecture built for governance and observability. Source: aws.amazon.com

Responsible Agentic AI Playbook targets public-sector systems

AI is increasingly embedded in public systems that shape economic mobility, including health, housing, workforce and benefits access. The playbook addresses how agentic AI should be deployed responsibly in them. Source: urban.org

Airline bans AI agents from booking flights on customers' behalf

Some airlines still allow AI agents to book flights, but consumer advocates warn of serious drawbacks. One carrier is now prohibiting the practice outright. Source: travel.yahoo.com

More AI Agents & Automation news

Advocacy groups sue AI leaders to freeze new models, citing 'catastrophic' risks

A coalition of online safety and consumer advocacy groups sued five of the world's top AI companies on Thursday. They want the courts to stop further development of new models, warning of catastrophic risks. Source: miamitimesonline.com

Nonprofits sue five AI giants, including Anthropic and Google, to freeze the AI race

A lawsuit filed in San Francisco targets five tech companies and asks a court to block the development of more advanced and self-improving AI models. The plaintiffs are nonprofit groups seeking to halt the escalating race among leading AI developers. Source: politico.com

US nonprofits take tech giants to court to stop AI development

A coalition of digital safety and consumer watchdog groups has filed a lawsuit against the world's leading AI developers. The suit asks the court to halt further development of the technology. Source: aljazeera.com

AI safety groups sue five tech giants, demand halt to new model development

The lawsuit argues that the companies are building increasingly autonomous systems even though they acknowledge the potentially catastrophic risks involved. Source: nationalheraldindia.com

'An existential threat': Pioneer Press workers fight for AI protections

Unionized workers at St. Paul's Pioneer Press and Adtaxi want AI provisions written into their next union contract. They are also appealing to readers for support. Source: workdaymagazine.org

Anthropic says its RSP trigger has been met: what the ASL levels mean

Anthropic states that the trigger in its Responsible Scaling Policy has been reached. The piece explains what the ASL threshold means, which safeguards apply, and how this affects Claude. Source: shattered.io

More AI Safety & Alignment news

Microsoft outage hits Copilot and 10 other 365 services

Microsoft said on its service health status page on 12 November that Microsoft 365 users may be unable to connect to multiple services. The list of affected services includes Copilot. Source: fourweekmba.com

Microsoft outage hits 365, Teams, Outlook and Copilot

Users reported problems accessing Microsoft 365, Azure, Outlook, Teams and other services on Thursday, Nov. 12. Source: mashable.com

Build once, run on any agent: a practical guide to agent skills

Agent skills capture repeatable engineering work in files that teams can review and reuse. Tools and permissions, however, stay specific to the host environment. Source: github.blog

Microsoft Copilot AI picks every NFL Week 10 game

Copilot had a near-perfect record with its NFL picks last week. Now the chatbot makes its predictions for Week 10. Source: usatoday.com

llm 0.37.2 ships with a security fix

A new version of llm, the command-line tool for accessing large language models, has been released as 0.37.2. The update patches a security vulnerability reported by Daniel de Araujo. Source: simonwillison.net

Microsoft scrambles to fix massive Copilot and Teams outage

Microsoft Copilot, Teams and other products and services went down on Thursday in a widespread outage. The company is working on a fix. Source: crn.com

Microsoft racing to fix Outlook and Copilot outage after nine hours

Microsoft is struggling to restore access to some of its flagship software more than nine hours after the outage began. Users remain unable to access the affected services. Source: bloomberg.com

Microsoft wants developers to spend less time fixing AI mistakes

The updated Copilot experience solved roughly 20% more coding tasks in Microsoft's testing. It also delivers shorter, more focused responses. Source: windowscentral.com

Microsoft outage knocks out Outlook, Teams and Copilot worldwide

Microsoft said it is working to restore access to its workplace software after customers reported trouble connecting to email and other services. Users in any region were affected. Source: finance.yahoo.com

How Microsoft made Copilot Cowork adoption stick internally

Microsoft shares its practical approach to encouraging employees to use Copilot Cowork. The article describes lessons learned after launching the tool inside the company. Source: microsoft.com

VS Code 1.146 brings AI agent inbox, model comparisons and chat windows

An experimental Inbox mode organizes AI agent sessions across projects and flags those that need attention. Model comparisons now also support Copilot Fusion. Source: visualstudiomagazine.com

More AI Tools & Products news

Study of 2,212 Viral AI Prompts Reveals What Works

An analysis of 2,212 shared AI video and image prompts looks at their median length, use of timecodes and exclusions. It also examines which traits tend to earn the most likes. Source: hackernoon.com

Top 5 AI Anime Video Generators of 2026 — Worth the Effort?

AI video tools can now turn a text prompt or a character image into a few seconds of animation, and anime is among the most requested styles. The article ranks five generators and asks whether AI anime clips are worth making. Source: sitepoint.com

9 Best Suno Music Video Generator Tools for 2026

Tracks made in Suno get little traction on YouTube or TikTok when posted as a static audio link. The article rounds up nine tools that turn Suno songs into music videos. Source: newwavemagazine.com

APIXO AI Adds Seedream 5.5 and Cuts API Prices by 30%

Seedream 5.5 brings image generation, editing and layer separation to the APIXO AI platform. Black Friday API pricing starts at $0.014 per 1K text-to-image output. Source: knoxnews.com

AI-Generated Adult Content: How It Works and What Risks It Carries

AI-generated NSFW content attracts a global audience of hobbyists, creators and people seeking tailored adult material via text-to-image generation. The piece explains the phenomenon and the risks involved. Source: eu-opensci.org

More Image & Video Generation news

Figure AI IPO: what we know about timing, valuation and how to invest

Figure AI, which builds general-purpose humanoid robots for warehouses, factories and eventually homes, is still privately held. This explainer covers what investors should know about a possible listing. Source: smartasset.com

Cathie Wood: Humanoid robots a $26 trillion market within 15 years

ARK Invest's Cathie Wood says humanoid robots could become a $26 trillion market over the next 10 to 15 years. She argues that 'embodied AI' could eventually account for roughly 20% of global GDP. Source: finance.yahoo.com

J.P. Morgan: US humanoid output could reach 400,000 units by 2030

J.P. Morgan forecasts that US humanoid robot production will reach 300,000 to 400,000 units by 2030 and about 1.5 million units by 2035. The bank expects deployments to ramp up over that period. Source: reuters.com

Agility Robotics goes public as first U.S. pure-play humanoid maker

Agility Robotics has entered the public markets, backed by more than $300 million in conditional multiyear orders for its Digit 5 humanoid robot. It is billed as the first U.S. publicly listed company specializing in humanoids. Source: techtarget.com

More Robotics & Embodied AI news

Yann LeCun launches AMI Labs in Paris after leaving Meta

Turing Award winner Yann LeCun has founded AMI Labs in Paris following his departure from Meta. He serves as founder and executive chairman of the new venture, which champions the JEPA world-model approach to AI. Source: finance.biggo.com

ML and Deep Learning Speed Up Antimicrobial Resistance Prediction

A scientific review describes how machine learning and deep learning models can rapidly predict antimicrobial resistance using multi-omic datasets. The approach promises faster insight into resistance patterns than traditional methods. Source: bioengineer.org

New Research Probes Theoretical Limits of LM Alignment

Language model (LM) alignment aims to steer model outputs toward human preferences without sacrificing the base model's capabilities. The paper examines the theoretical boundaries of how far this approach can go. Source: machinelearning.apple.com

Google unveils Gemini 4 Hydrogen, a fast and cheap model for AI agents

Google has announced Gemini 4 Hydrogen, a model designed to be quick and cost-efficient while running agentic workflows and routine everyday tasks reliably. The release targets developers who need dependable performance without the expense of larger models. Source: deepmind.google

More AI Research news

Databento Secures $100 Million Series C at $1.1 Billion Valuation

Salt Lake City-based Databento, a market data platform for modern finance, announced a $100 million Series C round. The financing values the company at $1.1 billion. Source: hjnews.com

Mirendil Lands $1 Billion to Build AI With Leaner Human Teams

Mirendil, a startup founded by Anthropic alumni, has raised $1 billion to develop AI faster and with fewer people involved. The round arrives just six months after the company announced a $200 million seed round. Source: wsj.com

Figure AI IPO Guide: Timeline, Valuation and How to Invest

Figure AI builds general-purpose humanoid robots aimed at warehouses and factories, with homes as a longer-term goal. The company is still private, and the piece outlines what investors should know about a potential IPO. Source: smartasset.com

More AI Business & Funding news

Huang's self-inflicted turmoil reportedly pushes Nvidia into a midlife crisis

One opinion piece claims Nvidia founder and CEO Jensen Huang has been ousted after almost two weeks of controversy that played out more like a soap opera than real corporate life. The snippet is cut off and gives no further details, so the claim should be treated with caution. Source: theregister.com

Nvidia faces life after Jensen Huang

An article claims Jensen Huang is no longer CEO of Nvidia. The authors say they believe the departure is directly linked to large-scale smuggling of GPUs, though the snippet offers no evidence for this. Source: newsletter.semianalysis.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 ☕