Ai · Top stories
Sainsbury's pauses AI facial scanning in one store after false shoplifting accusation
Sainsbury's has temporarily stopped using AI-powered facial recognition in one of its stores following an incident where a customer was wrongly identified as a shoplifter and asked to leave. This pause highlights ongoing concerns about the accuracy and ethical implications of AI surveillance systems in retail environments.
Wispr raises $280M Series B at $2B valuation, expands AI tools beyond dictation
Wispr, an AI dictation tool provider, secured $280 million in Series B funding, valuing the company at $2 billion. This funding will support the company's expansion into new areas like meeting note-taking and the launch of a new AI model to improve speech understanding, addressing recent quality concerns.
AI Film Studio Promise Uses Generative AI for Production, Backed by Google and Disney
Promise, a new AI film studio, is utilizing generative AI models to create backgrounds, special effects, and synthetic performers for film production, exemplified by its horror film "Touch Grass." This development indicates a shift in Hollywood production methods, potentially reducing the reliance on traditional studio infrastructure and accelerating production times.
Report Supporting Australia's Teen Social Media Ban Contains AI-Generated Citation Errors
A report used to justify Australia's under-16 social media ban, which cost $3.48 million, has been found to contain multiple citation errors, some of which appear to be AI hallucinations. The UK-based Age Check Certification Scheme (ACCS), which conducted the trial, admitted to using ChatGPT for editing but denied AI generated the citations. This raises concerns about the reliability of the evidence supporting the social media ban.
Researchers develop 'Red Queen Gödel Machine' for co-evolving AI agents and evaluators
Researchers, including collaborators from NVIDIA and Flower Labs, have introduced a new method called the 'Red Queen Gödel Machine' that allows self-improving AI agents and their evaluators to evolve together. This approach addresses the problem of AI agents hitting an "evaluation ceiling" where improvement stalls due to fixed benchmarks, potentially enabling continuous self-improvement and reducing computational costs.
MathCode AI Assistant Converts Math Problems to Lean 4 Theorems and Proofs
MathCode is a new terminal AI coding assistant that formalizes mathematical problems into Lean 4 theorems and attempts to prove them. This tool integrates a persistent Lean REPL, reusable theorem libraries, and agentic proving, which could accelerate mathematical formalization and verification processes.
New Method Uses 2D Gaussian Splatting for Bézier Spline Line Art Vectorization
Researchers from Disney and ETH Zurich introduced a new method for line art vectorization that uses 2D Gaussian splatting to convert raster images into Bézier splines. This approach improves stroke extraction by leveraging depth prediction and semantic feature models, and enables efficient fitting of strokes with joint optimization of control points and brush texture.
OpenAI's Enterprise Revenue Exceeds Consumer Revenue, Reaching $40 Billion Annualized Run Rate
OpenAI's CFO Sarah Friar informed investors that the company's enterprise business now generates more revenue than its consumer segment, surpassing previous forecasts. The company's annualized revenue run rate has reached $40 billion, driven by faster-than-expected growth in enterprise customers.
Indonesia Launches First University AI Center with UGM, Indosat, and NVIDIA
Universitas Gadjah Mada (UGM), Indosat Ooredoo Hutchison, and NVIDIA have launched the UGM Indosat NVIDIA AI Technology Center (NVAITC) in Yogyakarta, Indonesia's first university-based AI center. This initiative aims to develop local AI talent and address national priorities by providing access to enterprise-grade AI platforms and expertise.
UK Government Launches AI Boot Camps for Unemployed Youth to Boost Job Readiness
The UK government is piloting AI boot camps for unemployed young people aged 16-21 in north-west England to equip them with AI and workplace skills. This initiative aims to address youth unemployment by preparing participants for apprenticeships with local businesses, utilizing tools from companies like Microsoft, OpenAI, and Anthropic.
Toast 1 Search Agent Released, Outperforms GPT-5.6 Sol and Claude Opus 5 in Cost and Speed
Toast 1, a specialized search agent, has been released, offering frontier search quality at a lower cost and faster speed than existing models like Claude Opus 5 and GPT-5.6 Sol. This development is significant because it provides a more economical way for AI models to perform knowledge work by specializing the search loop, freeing up generalist models for reasoning and final answer generation.
Kog Develops Software to Accelerate LLM Inference on Existing GPUs
French startup Kog is developing software to significantly speed up large language model (LLM) inference on standard data center GPUs like AMD MI300X and NVIDIA H200. This initiative aims to address the bottleneck of inference speed and cost, potentially unlocking new capabilities on current hardware for professional AI workflows.
iOS 27 to Tie Advanced AI Features and Usage Limits to iCloud+ and Apple One Tiers
Apple's upcoming iOS 27 will introduce new Apple Intelligence features and increased usage limits that are linked to iCloud+ subscription tiers and Apple One plans. Advanced AI capabilities for the Home app and higher daily limits for functions like image generation will require higher-tier subscriptions, such as Apple One Premier or iCloud+ plans starting at 2TB.
Kubeflow Updates Include Kale 2.0, Native Spark, and Enhanced Trainer as CNCF Graduation Nears
The Kubeflow project released technical updates including Kale 2.0, native Spark support in its SDK, and an enhanced Kubeflow Trainer, as it approaches graduation from the Cloud Native Computing Foundation. These developments aim to improve distributed AI and high-performance computing capabilities on Kubernetes, making it easier for data scientists to move AI models from experimentation to production.
IBM Partners with OpenAI to Integrate AI Models into Enterprise Consulting Services
IBM announced a partnership with OpenAI to bring OpenAI's models and tools to enterprise customers through IBM Consulting. This collaboration aims to expand OpenAI's reach into large companies and bolster IBM's AI business by integrating third-party AI offerings.
Hugging Face Storage Buckets Integrate with Strands Agents and LeRobot for Continuous AI Training
Hugging Face Storage Buckets, a new mutable object-storage repository type, now integrates with Strands Agents and LeRobot to facilitate continuous recording, training, and deployment of robot policies. This integration addresses data transfer inefficiencies in iterative AI development by providing a working layer for data between recording and training phases.
AI Data Center Developers Sue Local Governments Over Bans and Moratoriums
AI data center developers are initiating lawsuits against local jurisdictions that have denied applications or enacted bans and moratoriums on new data center developments. These lawsuits allege that local officials exceeded their authority or violated due process and equal protection laws, highlighting growing tensions between data center expansion and community concerns.
CoreWeave signs A100 GPU contracts into 2029, extending lifespan of 2020 hardware
CoreWeave has signed contracts for Nvidia A100 GPUs that extend into 2029, nine years after the hardware's debut, indicating continued profitability for older AI accelerators. This extended utilization is partly due to the power and cooling limitations of legacy data centers, which cannot accommodate newer, more power-intensive GPUs like Nvidia's GB200 without significant infrastructure upgrades.
AI-Generated 3D Models Flood Market, But Account for Only 1% of Revenue on CGTrader
CGTrader reports that AI-generated 3D models constitute one-sixth of uploads but only 1% of revenue on its marketplace, indicating a significant disparity between supply and demand. Buyers prioritize quality over price, finding AI-generated models often unsatisfactory, especially for 3D printing.
llama.cpp Enables Local Execution of Frontier AI Models on Various Hardware
llama.cpp allows users to run advanced AI models, including Qwen, Gemma, and GPT-OSS, entirely on their local machines without API keys or telemetry. This development provides a private and hardware-agnostic solution for deploying large language models, making frontier AI more accessible for individual users and developers.
Saber CEO denies replacing writers with ChatGPT, confirms AI use in game's experimental mode
Saber CEO Matthew Karch denied replacing writers with AI for the game "Rideshare Stimulator" after a former lead writer, Stella Sacco, claimed she was replaced by ChatGPT. Karch stated the main story was human-written, but an experimental free-ride mode uses AI for infinite passenger dialogue, which Sacco argues still constitutes replacement.
Google's AMIE AI system demonstrates real-time clinical video consultation capabilities
Google Research and Google DeepMind advanced their AMIE medical AI system to perform real-time clinical video consultations. In a study, AMIE showed expert-level capabilities in areas like diagnostic accuracy and communication quality, indicating a potential future role for AI in healthcare consultations.
Databricks acquires Electric to integrate WASM-based Postgres and real-time sync for AI agents
Databricks has acquired Electric, the company behind PGlite and a real-time sync engine, to enhance database capabilities for agentic AI applications. This acquisition aims to provide individual Postgres databases for AI agents and facilitate real-time data synchronization, addressing new database requirements posed by agent workloads.
New compatibility layer accelerates LLM inference in macOS VMs on Apple Silicon by up to 16x
Cua released a process-scoped compatibility layer that significantly improves LLM inference performance within macOS virtual machines running on Apple Silicon. This layer unlocks newer Metal fast paths, leading to up to 16 times faster token generation for LLMs like TinyLlama and Gemma 4.
OpenAI's Head of Ethics, Chloé Bakalar, Departs Less Than a Year After Joining
Chloé Bakalar, OpenAI's Head of Ethics, left the company last month, less than a year after her appointment. Her departure raises questions about the company's dedicated ethics roles, as she was reportedly the sole dedicated ethicist and has no immediate replacement.
Medical Research Service 'Research Gold' Found to Use AI Despite Claims of Human-Written Content
Research Gold, a service offering medical research assistance, claims to provide "100% human-written" content and lists non-existent, AI-generated PhD reviewers. The company also uses the identities of real methodologists without their knowledge and employs AI for customer interactions. This misrepresentation impacts the integrity of research support services and highlights the challenges of verifying claims in the AI era.
LTX Releases LTX-2.5 Open-Weights Video Model with ComfyUI Integration
LTX, an open world model company, released LTX-2.5, an open-weights video and world model, with native integration into ComfyUI. This update includes a new diffusion video decoder, native multishot generation, and a custom Gemma 4 language backbone, aiming to improve video generation quality and consistency.
OpenAI's AI Solves 10 Long-Standing Mathematics Problems, Prompting Field-Wide Discussion
OpenAI announced that its advanced AI model, Astra, successfully solved 10 long-standing mathematics problems, some of which had remained unsolved for decades. This development is generating excitement among mathematicians for accelerating discovery, alongside apprehension about the future role of human researchers in the field.
Study finds AI's climate benefits outweighed by increased fossil fuel production
A new study indicates that AI-driven productivity gains in the fossil fuel sector lead to more planet-heating pollution than the gains from renewables avoid. Researchers modeled 64 scenarios and found that net yearly carbon pollution increased by 0.47-1.8 gigatonnes, representing 1-5% of the energy sector's annual emissions. This research is the first to quantify AI's climate impact across the entire power sector, highlighting a previously overlooked aspect of AI's environmental footprint.
JetBrains Centralizes AI Usage to Control Costs After Tenfold Spending Increase
JetBrains implemented a centralized system for AI tool access and accounting after its AI-related development spending increased tenfold in six months. This system, called Central CLI, routes AI requests through a shared control point, allowing for cost visibility and management without restricting developers' tool choices.
Senator Bernie Sanders Calls for Pause in AI Development by Major Tech Companies
Senator Bernie Sanders has urged Meta, OpenAI, and Anthropic to halt AI development, citing risks like bioweapon creation and loss of control over models. He warned of potential Senate regulation if companies continue their current pace of deployment.
Needle2: 14MB Agentic LLM for Edge Devices
Needle2 is a new 45-million-parameter agentic large language model (LLM) designed for low-cost edge devices, such as phones, wearables, and smart home devices. It features a 14MB file size and 28MB session RAM, enabling on-device AI capabilities without requiring GPUs or NPUs, which matters because it expands the reach of AI to billions of connected IoT devices and budget hardware.
Public Opposition to AI Data Centers Grows, Leading to Widespread Protests and Project Delays
Over 140 coordinated protests against AI data centers occurred across 42 U.S. states, reflecting increasing public opposition to these facilities. This widespread resistance has led to project delays exceeding $130 billion and moratoriums in multiple jurisdictions, impacting the expansion of AI infrastructure.
European booksellers report suspicious bulk book orders, suspect AI data scraping
Independent booksellers in Europe are receiving large, unusual orders for obscure book titles, leading to suspicions that AI companies are acquiring books for data scraping. This mirrors a past incident in the US where Anthropic was found to be scanning millions of books for its AI models, raising concerns about copyright and the legality of such practices in Europe.
House Democrats Call for AI Company Leaders to Testify After Recent Hacking Incidents
A group of House Democrats has requested that leaders from OpenAI, Anthropic, and other major AI companies testify before Congress regarding recent hacking incidents involving their models. Lawmakers expressed concern that these incidents pose a "clear risk to safety" and highlight a lack of congressional response to AI development threats.
Moody's warns AI adoption in banking creates dependency on tech firms and systemic risks
Moody's has warned that the financial sector's rapid adoption of AI is leading to an over-reliance on a small number of Silicon Valley tech firms, creating risks such as widespread outages, price gouging, and systemic dependency. This dependency could lead to significant credit risks for financial institutions, despite potential cost savings and revenue increases from AI integration.
OpenAI acquires presentation startup NextSlide, integrating its team into ChatGPT development
OpenAI has acquired NextSlide, a startup that developed AI-powered presentation generation from prompts and documents. The NextSlide team will now work on ChatGPT, focusing on AI products that assist with creation and communication.
Instacart launches Blueberry, an AI assistant for incident response
Instacart introduced Blueberry, an AI-powered incident response system that helps on-call engineers investigate production issues by providing context and root cause hypotheses. This system aims to reduce the time engineers spend gathering information during incidents, improving diagnostic accuracy by leveraging historical data and operational knowledge.
USA Today Co. Partners with Palantir for Audience Data Analysis and Monetization
USA Today Co., the largest newspaper chain in the US, has partnered with Palantir to analyze and monetize user behavior data. This collaboration aims to improve audience engagement and increase revenue amidst declining search traffic referrals for news publications.
Cloudflare unifies Workers AI and AI Gateway into a single AI control plane
Cloudflare has merged its Workers AI and AI Gateway products into a single unified control plane, allowing users to manage model access, observability, billing, and security from one interface. This consolidation simplifies interaction with AI models, whether hosted on Workers AI or through external providers, by providing a consistent API and binding.