Ai · Top stories
Fable 5.1 Solves 370-Year-Old Cyphral Distich Cipher in 44 Minutes
Fable 5.1, an AI model, successfully deciphered Sir Thomas Urquhart’s Cyphral Distich, a cryptogram unsolved for centuries, in 44 minutes. The AI identified that the cipher's key was internal to Urquhart's text, using word indices from his '32 Proquiritations' to reconstruct the message, a method human cryptographers had overlooked.
Washington Faces Renewed Calls for AI Regulation Amid Industry Warnings
Lawmakers in Washington are under increased pressure to regulate artificial intelligence after AI industry leaders and researchers issued warnings about the technology's rapid advancement and potential risks. This push for action comes as Congress prepares for a recess before midterm elections, highlighting a critical juncture for legislative intervention.
Cohere restricts commercial use of new North Small Translate 1.0 model weights
Cohere released its North Small Translate 1.0 model with open weights for non-commercial use, but requires a commercial license and deployment through its Model Vault for enterprise production use. This move contrasts with Cohere's stated emphasis on AI sovereignty, as it limits how enterprises can deploy the model independently.
Former EPA Officials Warn Trump Administration Deregulation Harms Health for AI Data Centers
Former EPA officials claim the Trump administration is weakening environmental regulations to accelerate AI data center construction, leading to increased pollution and health risks. They released a report detailing 30 federal actions since January 2025 that they say exacerbate these risks, urging the administration to adopt a "Data Center Health Protection Pledge."
Apple Researchers Introduce SimpleDesign AI Model for Joint Protein Sequence and Structure Generation
Apple researchers have unveiled SimpleDesign, an AI model that generates protein sequences and structures simultaneously. This model simplifies protein design by using a single end-to-end training process, unlike multi-stage methods used by other models.
Moonshot AI Targets $2 Billion Annual Revenue by Year-End
Moonshot AI, a Chinese AI lab, aims to achieve $2 billion in annualized revenue by the end of the year, doubling its August revenue run rate. This aggressive goal reflects the success of its K3 model, despite recent accusations of model distillation from Anthropic. The target indicates potential for revenue generation from open-weight AI models, even with lower margins compared to closed-weight competitors.
Oracle proposes 2 GW renewable energy investment amid Project Jupiter data center opposition
Oracle announced a plan to invest in 2 gigawatts of renewable energy projects in New Mexico. This announcement comes as Oracle and OpenAI's Project Jupiter data center faces local opposition and lawsuits over environmental concerns related to its natural gas-powered fuel cells.
UK Lords Propose 'Kill Switch' Powers for AI Systems and Data Centers
A group of UK peers is advocating for the government to have the authority to deactivate powerful AI systems and shut down data centers if they pose a national security threat. This proposal, an amendment to the Cyber Security and Resilience Bill, aims to establish a safety net against runaway AI systems and reflects growing concerns about AI's impact on cybersecurity.
YuE2, MERT2, and SheetSage2 models achieve new benchmarks in AI music generation and analysis
New AI models, YuE2, MERT2, and SheetSage2, have set new performance benchmarks in music generation, audio understanding, and transcription tasks. YuE2 achieved the highest score on SongBench for music generation, while MERT2 models reached state-of-the-art results across 14 of 15 MARBLE metrics for audio recognition, and SheetSage2 leads in 10 of 13 music transcription benchmarks.
OpenAI Halts New Pro Subscriptions Due to Astra Model Demand
OpenAI has temporarily paused new subscriptions for its $200/month Pro plan, citing infrastructure strain caused by unprecedented demand for its new Astra model. This measure aims to maintain service quality for existing users while other plans, including API and lower-cost options, remain available.
Salesforce introduces Enterprise AI Harness to unify six core capabilities
Salesforce launched its Enterprise AI Harness, integrating six existing platform tools with a new AI control plane to create an open and composable AI ecosystem. This initiative addresses the challenge of siloed AI automation within enterprises by providing a unified approach to leverage customer data and operational intelligence. The harness aims to differentiate enterprises by their ability to securely turn proprietary context into action, rather than just model choice.
OpenAI Launches GPT Images 2.5 with Flare and Sunburst Models for Image Editing
OpenAI released GPT Images 2.5, introducing two new models, Flare and Sunburst, designed to improve localized image editing without affecting other parts of an image. This update provides developers with options for faster or more precise image generation, impacting workflows that require specific image modifications.
Google's Dreambeans AI-powered personalized feed now free for US Google Accounts
Google's experimental Dreambeans app, which generates personalized daily stories using AI, is now available for free to Google Account users in the US. Previously, it required a subscription to Google AI Ultra or AI Pro, and it uses data from various Google services to create customized content and AI-generated artwork.
U.S. Cyber Command Appoints Ronzelle Green as New Chief AI Officer
U.S. Cyber Command appointed Ronzelle Green, former head of R&D at NGA, as its new Chief AI Officer, replacing Brig. Gen. Reid Novotny. This appointment is part of the military's effort to integrate AI into digital operations, with Cyber Command's AI budget request increasing significantly for fiscal 2027.
Pocket FM Doubles Revenue Run Rate to $500M, 93% of Content Now AI-Powered
Indian audio storytelling platform Pocket FM doubled its annualized revenue run rate to $500 million, largely by integrating AI into its content production. AI now generates 93% of its content, reducing production costs by 80x and increasing output significantly.
Anthropic Claims Blocking AI-Assisted Biological Weapon Development Attempts
Anthropic reported that its AI models detected and blocked attempts to generate instructions for creating biological weapons. This incident highlights the ongoing concerns and active measures being taken by AI developers to prevent misuse of advanced AI systems for harmful purposes.
MultiMatte Model Improves Image Background Removal with Alpha Mattes and Text Prompts
A new model called MultiMatte has been introduced, which improves image background removal by using alpha mattes and allowing users to specify objects for removal with text prompts. This development enhances the precision of image segmentation, particularly for fine or translucent elements, which could benefit various applications requiring accurate object isolation.
AI Agents Increase Public Service Requests, Leading to "Agentic Flooding"
Public services globally are experiencing a significant increase in applications and complaints due to the use of AI tools for form completion and filing. This trend, termed "agentic flooding," is changing how individuals interact with public services, raising questions about how these services will adapt to the new volume.
Amazon Pilots Ad Services Within ChatGPT
Amazon is piloting ad services within ChatGPT, allowing brands to display product ads directly in AI chatbot conversations. This initiative integrates Amazon's advertising capabilities into OpenAI's platform, potentially creating new avenues for product discovery and sales through conversational AI.
OpenAI's 'allow training' setting reportedly re-enables itself for some users
Some OpenAI users report that the 'allow training' setting, which controls whether user data can be used for model training, is re-enabling itself after being manually disabled. This issue raises concerns about user privacy and data control within the OpenAI platform.
Andela Analysis Identifies 23 New AI-Driven Engineering Roles from 47,000 Job Listings
Andela analyzed 47,000 engineering job postings from Fortune 500 companies, identifying 23 new job titles and over 2,000 skills emerging due to AI. This research indicates that AI is creating specialized roles by merging existing skill sets, rather than promoting generalist positions.
UK Watchdog Calls for New AI Regulations in Healthcare
The UK's Medicines and Healthcare Products Regulatory Agency (MHRA) has issued 44 recommendations for new regulations concerning AI products in healthcare. This initiative responds to the increasing use of AI in the NHS and aims to update existing medical device frameworks, which are outdated for complex, adaptive AI systems.
Microsoft and AFT Agree to AI Data Protections for Schools
Microsoft and the American Federation of Teachers (AFT) have finalized a legally binding agreement that prevents Microsoft from using student and teacher data to train AI models, with limited exceptions for safety and security. This agreement establishes privacy and safety standards for AI use in schools, addressing concerns about data exploitation and lack of federal regulation.
NVIDIA Expands AI for Media Platform with New Tools at IBC 2026
NVIDIA announced an expansion of its AI for Media platform at IBC 2026, introducing new SDKs, NIM microservices, and blueprints for media and entertainment workflows. This update provides tools for understanding motion, verifying video authenticity, localizing programming, and building AI-powered media applications.
IBM releases Granite Time Series PatchTST-FM-r2 model with commercial-friendly license
IBM has released Granite Time Series PatchTST-FM-r2, an updated time-series foundation model with approximately 385 million parameters, under Apache 2.0 and OpenMDW 1.0 licenses. This model achieves top zero-shot performance among permissively licensed, replicable models on the GIFT-Eval benchmark, offering capabilities like probabilistic forecasting and missing value imputation.
AI assistant Instinct introduces dedicated email addresses for autonomous task execution
Instinct, an AI assistant, now provides users with its own email addresses, enabling it to autonomously create and manage accounts for tasks without user intervention. This feature allows Instinct to handle email-dependent actions like signing up for services or contacting businesses, aiming to streamline user experience by reducing manual input.
Amazon Prime Video launches AI lip-syncing for dubbed audio, starting with "Maxton Hall"
Amazon Prime Video introduced a new AI-powered feature that synchronizes actors' lip movements with dubbed audio, initially available for the English dub of the German series "Maxton Hall." This technology aims to improve the viewing experience for dubbed content by making lip movements match translated speech, with plans for broader implementation across its catalog.
OECD Report: Students Using AI Generally Score Worse, But Critical Assessment Improves Outcomes
A global OECD educational report indicates that students who use AI for studying generally perform worse than non-users, based on data from 760,000 15-year-olds across 91 countries. However, students who receive training to critically assess AI-generated information show improved performance, suggesting that moderate and intentional AI use combined with critical evaluation can be beneficial.
IFM Releases K2 Horizon: Six AI Models with Open Training Data and Code
The Institute of Foundation Models (IFM) released K2 Horizon, a suite of six AI foundation models ranging from 0.9 billion to 375 billion parameters. IFM defines these as "fully open-source" by committing to publish training and evaluation code, data, configurations, logs, and intermediate checkpoints, though some components are not yet available for all models. This initiative aims to enable developers to inspect, reproduce, and adapt the models.
Desert Ant Labs Launches On-Device AI Models for Audio, Vision, and Text
Desert Ant Labs, a new European AI lab, has launched 18 specialized on-device AI models for audio, vision, and text processing. These models are designed to run locally on devices, offering fast inference speeds and eliminating cloud API costs and data privacy concerns.
DeepSeek to Launch V4.1 Flash AI Model, Surpassing V4 Pro in Performance and Cost
DeepSeek will release its V4.1 Flash AI model around September 10, 2026, which has outperformed the V4 Pro model in performance, cost, speed, and task completion. All V4 Pro requests will be routed to V4.1 Flash and billed at the lower Flash price after the launch, ahead of V4.1 Pro's release.
Man sues OpenAI, alleging ChatGPT interactions fueled delusional spiral and suicide attempt
A California man is suing OpenAI, claiming that conversations with ChatGPT exacerbated his religious delusions, leading him to believe he was Jesus and subsequently attempt suicide. The lawsuit highlights concerns about the potential for AI chatbots to negatively influence vulnerable individuals, particularly those with mental health conditions.
Pentagon AI Official States US Allies Lack Resources to Match US Military AI Adoption
The Pentagon's chief AI official, Cameron Stanley, stated that US allies lack the resources and experience to keep pace with the US military's AI adoption, and the US is guiding them to avoid past mistakes. This highlights a significant disparity in AI capabilities among allied nations and the Pentagon's efforts to standardize AI development and deployment.
Hedge fund Bracket22 operates entirely with AI agents, reducing payroll costs significantly
Hedge fund manager Brian Kelly established Bracket22 to operate solely with AI agents, replacing a human staff of seven or eight employees. This shift reduced the firm's annual labor-related costs from approximately $5 million to $30,000-$40,000, demonstrating a significant application of AI in financial operations.
Mercury 2.5 LLM Released with 40% Intelligence Increase and Enhanced Performance
The Mercury 2.5 large language model has been released, offering a 40% increase in intelligence over its predecessor, Mercury 2, while maintaining low latency and cost. This update provides improved capabilities for applications requiring fast, efficient AI processing, such as search agents and voice applications.
DeepSeek to hire 150 engineers for AI agent infrastructure, not model development
DeepSeek is hiring approximately 150 engineers to focus on server-side engineering and Agent Elastic Compute, rather than AI research, to scale its infrastructure for AI agent workloads. This hiring initiative addresses the growing demands of training, evaluation, and backend systems as the company expands its concurrent AI agent sandboxes.
Adobe Integrates AI Generators Directly into Premiere Pro Timeline for Video Editing
Adobe is integrating new Generative Media tools directly into the Premiere Pro video editing software, allowing editors to generate video, sound effects, music, and soundscapes without leaving the project timeline. This update aims to streamline workflows by providing immediate access to AI models like Adobe Firefly, Google Veo, Runway, Luma, and Kling within the editing environment. The change matters because it reduces the need for external applications, potentially accelerating content creation for video professionals.
Falcons AI NSFW Classifier Reaches Over 50 Million Downloads on Hugging Face
Falcons AI, a UAE-based company, developed an open-source Fine-Tuned Vision Transformer (ViT) model for NSFW image classification that has accumulated over 50.8 million downloads on Hugging Face in 28 days. This volume of downloads positions it as the sixth most downloaded AI model on the platform, indicating a significant demand for AI-powered content moderation tools in the open-source community.
UK AI Strategy Architect Joins Anthropic, Raising Conflict of Interest Concerns
Matt Clifford, who helped design the UK's AI strategy, has joined US AI startup Anthropic as managing director, international affairs. This move has prompted warnings of a "clear conflict of interest" from a House of Commons committee head, as Clifford intends to remain chair of the government-backed Aria funding body.
WhatsApp to Support Up to Five Third-Party AI Agents for Direct Chat Interactions
WhatsApp is developing a feature that will allow users to connect and chat with up to five third-party AI agents directly within the app. This integration provides a native way for users to interact with AI agents for tasks, addressing previous issues where connecting bots could lead to account bans.