145 topics · 4 multi-source · newest first
OpenAI published a system card for GPT-5.5, detailing the model's capabilities and safety properties.
OpenAI launches GPT-Image-2, a new image generation model with improved text rendering and thinking capabilities, now available on API and ChatGPT.
NVIDIA outlines expanded collaborations with Adobe and WPP to deploy enterprise-grade, policy-governed AI agents for marketing and creative workflows, with security and governance features.
Anthropic demonstrates automated AI agents can conduct alignment research autonomously, outperforming human researchers on weak-to-strong supervision tasks.
Latent Space newsletter covers Claude Opus 4.7 launch, computer-use agents, and open-source agent research developments.
Anthropic launched Claude Opus 4.7, claiming improvements across reasoning, coding, vision, and instruction-following with unchanged pricing.
OpenAI announced workspace agents in ChatGPT, cloud-based tools designed to automate complex workflows and help teams coordinate work across multiple applications.
GitHub is stopping new registrations for Copilot Pro, Pro+, and student plans, tightening usage quotas for personal plans, and removing Opus model from Pro; Pro users can upgrade to Pro+ for higher limits, with refunds available if limits cause issues before 2026-05-20.
Anthropic announces new Claude connectors enabling direct access to personal apps like Spotify, Uber Eats, and TurboTax within conversations.
LiteParse, an open-source project, now has a web-based version that extracts PDF text in the browser using PDF.js and optional OCR (Tesseract.js) with spatial parsing to maintain readable layouts, and includes a UI for text and JSON output.
The US considers China’s distillation attacks on AI IP a threat and is preparing a crackdown, while China calls the accusations slander.
Sierra announces the acquisition of Fragment, a YC-backed French startup.
Meta is planning to lay off approximately 8,000 employees in May, according to a memo from the company's chief people officer.
Discussion of Tim Cook's potential departure and its implications for Apple.
NVIDIA and OpenAI announce GPT-5.5-powered Codex running on GB200 NVL72 infrastructure with enterprise security features and a decade-long collaboration.
Anthropic is investigating unauthorized access to its Claude Mythos model, reportedly available to a small group since the company announced controlled testing.
US government accuses Chinese actors of large-scale copying of American frontier AI models, with actions to counter.
New data-center natural gas projects in the US could emit up to 129 million tons of greenhouse gases annually, potentially rivaling Morocco's 2024 emissions.
Anthropic has restricted the viral OpenClaw AI agent tool and introduced paid access requirements to manage system strain and improve profitability.
Researchers caution policymakers may underestimate the implications of advancing world models and robotics, noting China’s progress in robotics.
Google reports that about 75% of its new code is generated by AI and then reviewed by human developers.
Microsoft is rolling out Agent Mode, an upgraded Copilot experience, in Word, Excel, and PowerPoint this week.
OpenAI announced automation capabilities for Codex enabling scheduled tasks and trigger-based workflows for reports and summaries.
OpenAI announced plugins and skills features enabling users to connect tools, access data, and automate workflows.
OpenAI announced a bug bounty program for GPT-5.5 offering up to $25,000 for identifying universal jailbreaks related to biosafety risks.
WIRED reports that Sam Altman’s Orb company promoted a Bruno Mars partnership that the artist says does not exist.
OpenAI released workspace agents in ChatGPT that can autonomously perform business tasks like gathering feedback and drafting emails for Business, Enterprise, Edu, and Teachers plan users.
SpaceX reportedly offered Cursor a $10 billion collaboration fee and $60 billion acquisition path, reportedly halting the startup's $2 billion fundraise.
MIT researchers introduce RLCR, a training method that calibrates AI confidence alongside answers, reducing calibration error without sacrificing accuracy.
Google Cloud launched two new TPU chips offering faster performance and lower costs than previous generations.
WIRED AI discusses AI models used to scam and their potential social dangers.
Google introduces TPU8t and TPU8i, claiming improvements for training and inference in an agent-era AI platform.
Google announced general availability of Gemini Embedding 2 through the Gemini API and Vertex AI.
Replit announces Auto-Protect to automatically detect, patch, and guide deployment of critical CVE-related fixes for dependencies.
A hacker group reportedly used AI to develop malware and operate fake websites, stealing about $12 million over three months.
OpenAI made ChatGPT free for verified U.S. physicians, nurse practitioners, and pharmacists to support clinical care and documentation.
NVIDIA and Google Cloud announce expanded cooperation to advance agentic and physical AI, including new Vera Rubin-based A5X instances, Gemini on Google Distributed Cloud, confidential computing, and open model tooling.
Google DeepMind published research on Decoupled DiLoCo, a technique for resilient distributed AI model training.
OpenAI introduced WebSocket support in the Responses API to reduce latency and API overhead for agentic workflows.
Cloudflare announces general availability of Sandboxes and Containers, offering persistent isolated Linux environments for AI agent workloads with features like credential injection, PTY support, and snapshot recovery.
Cloudflare proposes a reference architecture for scaling Model Context Protocol deployments with centralized governance, remote infrastructure, and cost controls.
OpenAI released an open-weight model designed to detect and redact personally identifiable information in text.
The US Defense Department requests about $53.6 billion in FY2027 for drone production, operations, logistics, and counter-drone systems under the Defense Autonomous Warfare Group.
Mozilla reports Mythos Preview identified 271 security vulnerabilities in Firefox 150 during early access, according to a company blog post.
Cybercriminals are increasingly using generative AI tools to scale phishing, deepfakes, malware, and fraud attacks, while AI-powered defenses also improve.
Multi-agent AI tools that coordinate multiple agents to complete complex tasks are beginning to ship from major AI companies.
Robotics companies are collecting human movement data at scale to train humanoid robots, raising questions about feasibility and labor implications.
Chinese AI labs are gaining market share through open-source model releases, reportedly surpassing US companies in global downloads.
AI companies develop autonomous research systems to assist scientific discovery, though research suggests potential risks to scientific diversity.
OpenAI released ChatGPT Images 2.0, a new image generation model that Sam Altman claims represents a leap equivalent to GPT-3 to GPT-5.
Florida is reportedly investigating potential liability of OpenAI regarding ChatGPT's role in a mass shooting, with officials citing chat logs; OpenAI denies responsibility.
Replit is awarded Google Cloud's 2026 AI Tooling Partner of the Year.
Replit unveils Security Agent to perform comprehensive security reviews of apps in under an hour using a hybrid Semgrep/HoundDog.ai approach.
A Google Research blog post discusses ReasoningBank, a system enabling agents to learn from experience.
Google announced Deep Research Max, an upgraded version of its autonomous research agent tool.
Anthropic announces Managed Agents, a layer for handling orchestration, sandboxing, and state management of Claude-based agent workflows.
OpenAI released ChatGPT Images 2.0, an updated image generation model with improved text rendering, multilingual support, and visual reasoning capabilities.
Editorial discusses the emergence of 'dark factories' and introduces Trycycle as a simple skill for planning and execution in AI-assisted software development, alongside related tooling and patterns.
DoorDash describes a hybrid approach using LLMs for consumer profiles and blueprints with deep learning for ranking to capture short-lived user intent.
Replit describes its defense-in-depth security architecture, grounded in Zero Trust, for securing customer code from development sandbox to production deployments.
Moonshot releases Kimi K2.6, a 1T-parameter open-weight MoE model reportedly competing with Opus and Gemini on coding and agentic tasks.
OpenAI launches Codex Labs and announces enterprise partnerships to scale Codex deployment across software development.
Git 2.54 adds an experimental git history command for simple rewrites, plus configuration-based hooks and default geometric repacking for maintenance.
Noetik signed a $50M deal with GSK to license AI models for cancer patient-treatment matching, aiming to improve clinical trial success rates.
LinkedIn announces Cognitive Memory Agent, a memory infrastructure layer for stateful, context-aware AI with persistent memory across episodic, semantic, and procedural layers.
Google introduces subagents in Gemini CLI to delegate tasks to specialized AI agents alongside a main session.
Google releases ADK for Java 1.0, adding external tools integration, app/plugin architecture, and context engineering features.
BAIR introduces GRASP, a gradient-based planner that enables long-horizon planning with learned world models by lifting dynamics into a parallelizable, collocation-based optimization framework.
NVIDIA and partners demonstrate AI-driven manufacturing capabilities and AI infrastructure at Hannover Messe 2026, highlighting AI-powered design, real-time simulation, and autonomous robotics.
Simon Willison discusses how headless APIs are becoming essential for personal AI agents to interact with enterprise services like Salesforce.
GitHub announces new status-page features: a Degraded Performance state, per-service uptime metrics, and a Copilot AI Model Providers component to improve incident transparency.
MIT awards the 2026 Harold E. Edgerton Faculty Achievement Award to Jacob Andreas and Brett McGuire for distinguished teaching, research, and service.
OpenProtein.AI provides a no-code platform and models for protein design, enabling biologists to use AI without deep ML expertise, with Boehringer Ingelheim expanding collaboration.
Simon Willison released llm-openai-via-codex, a plugin enabling LLM CLI access to GPT-5.5 via OpenAI's Codex subscription mechanism.
OpenAI announces GPT-5.5 described as autonomous agents able to switch tools, but details and claims are unverified.
A Stanford CS 153 event drew attention online and locally, eliciting mixed reactions.
Google has open-sourced the DESIGN.md prompt format used by its Stitch AI tool to help AI agents follow brand guidelines.
OpenAI and Microsoft are collaborating on a Trusted Access program to leverage capable AI models for cybersecurity defense.
Era secured $11M in funding to develop a software platform for diverse AI hardware form factors including glasses, rings, and pendants.
A Claude user survey finds new capabilities as the top productivity benefit, with speed close behind; creatives feel left behind, and the sample has notable bias.
Delve, a compliance startup, performed security certifications for Context AI before the AI training company disclosed a security incident.
A tech journalist argues the AI industry fundamentally misunderstands why people dislike AI, attributing resistance to a worldview gap rather than a marketing failure.
OpenAI released Privacy Filter, an open-source model that detects and redacts personal data in text.
Sony claims its table tennis robot Ace has reached expert-level performance in a sport.
GeForce NOW introduces in-app labels to show which titles are available from connected Xbox Game Pass and Ubisoft+ libraries, and adds six new games and a Thor skin.
OpenAI claims GPT-5.4-based ChatGPT for Clinicians outperforms doctors on clinical tasks even with unlimited time and web access, as reported by The Decoder.
OpenAI announced Codex, a tool designed to automate tasks, integrate with external tools, and generate outputs like documents and dashboards.
Google unveiled TPU v8 chips and Gemini Enterprise Agent Platform at Cloud Next, while OpenAI, Cursor, and others standardize agent harnesses for team workflows.
Tesla raised its 2026 capital expenditure plan to $25 billion, three times its historical spending, which the CFO said will result in negative free cash flow for the remainder of the year.
Google introduced new automated functions in Workspace powered by its Workspace Intelligence AI system.
X is replacing Communities with AI-powered custom timelines curated by Grok, including new advertising slots.
Sen. Elizabeth Warren warns that AI companies' massive spending and borrowing practices resemble pre-2008 financial crisis conditions and calls for congressional oversight.
Shopify CTO Mikhail Parakhin discusses the company's internal AI initiatives including Tangle, Tangent, and SimGym for ML experimentation and optimization.
Anthropic briefly tested removing Claude Code from the Pro plan for a small subset of new signups, later clarified as a small test.
Sony AI developed Ace, a robot that can beat elite table tennis players while following official ITTF rules.
Google has integrated Gemini-powered auto-browse capabilities into Chrome to help enterprise workers automate research and data entry tasks.
The article title suggests exploration of recomposing photos by changing angles using AI, but the content provided is minimal.
Google released Gemini Enterprise Agent Platform, an agent-building tool designed for IT and technical enterprise users.
Google is adding AI Overviews to Gmail for work accounts, providing instant summaries across multiple emails.
Alibaba's Qwen released Qwen3.6-27B, a 27-billion-parameter dense model claiming flagship-level coding performance in a 55.6GB package.
Google has released a Gemma 4 Vision Language Agent demo optimized for NVIDIA's Jetson Orin Nano Super edge device.
OpenAI and Infosys announced a partnership to integrate AI tools into Infosys's enterprise offerings for software development, modernization, and DevOps.
An Indian medical student describes earning money by creating and selling AI-generated bikini photos using a Gemini-based tool.
A NVIDIA Blog post outlines five ways NVIDIA AI aims to protect the planet, but no article body is provided.
Opinion piece argues that API design and tooling choices by model providers affect LLM reliability, not just the model itself.
A livestream announced by WIRED AI will address questions about the Musk v. Altman trial and its potential impact on OpenAI.
Pangram Labs released an updated Chrome extension that labels AI-generated content as users scroll social feeds.
Mozilla used an early version of Claude Mythos Preview to identify 271 vulnerabilities fixed in Firefox 150.
Anthropic reportedly tested restricting Claude Code to $100+/month plans before quickly reverting the change after public backlash.
MIT Technology Review explores emerging techniques to make large language models more efficient and capable of solving complex multi-step problems.
Researchers at Google DeepMind, Stanford, and other labs are advancing world models to help AI systems better understand and navigate physical environments.
Weaponized deepfakes are increasingly used for sexual exploitation, political propaganda, and disinformation, raising concerns about societal trust and election integrity.
ChatGPT Images 2.0 improves detail and text rendering but has language limitations.
Google has released Continued Conversation, a new feature for Gemini for Home designed to make interactions more natural and efficient.
Google DeepMind has partnered with global consultancies to help organizations adopt frontier AI technologies.
GitHub attributes recent outages to rapid growth, architectural coupling, and load-handling limitations.
Hugging Face launches QIMMA, a quality-focused Arabic language model leaderboard.
Hugging Face publishes commentary on the relationship between open AI development and cybersecurity practices.
Nathan Lambert analyzes why open models lag closed models and argues that benchmark scores increasingly misrepresent real-world AI capabilities.
Google AI Pro and Ultra subscribers now receive increased usage limits and expanded model access in Google AI Studio.
The piece argues for scenario planning to imagine divergent futures for AI and employment, citing mixed evidence on AI impact and outlining a two-axis framework to guide robust strategies across possible outcomes.
Simon Willison documents methods to fetch data from Datasette instances directly into Google Sheets using SQL functions.
Simon Willison upgraded his Claude Token Counter tool to compare tokenization across different Claude model versions.
Hyatt deploys ChatGPT Enterprise across its global workforce to improve productivity and guest experiences.
Simon Willison analyzes system prompt changes between Claude Opus 4.6 and 4.7, documenting updates to child safety, tool integrations, and response behavior.
Simon Willison documents Claude system prompt changes across versions using git timeline visualization.
GitHub demonstrates an emoji list generator runnable in the terminal using the Copilot CLI and Copilot SDK, with plans and multi-model workflow in plan mode.
Datasette 1.0a28 alpha release fixes compatibility bugs and adds automatic resource cleanup for database instances.
OpenAI published documentation on setting up and using Codex workspaces, threads, projects, and file management.
Noscroll has introduced an AI bot designed to read internet content automatically, aiming to reduce doomscrolling behavior.
A podcast crossover episode explores shifts in AI engineering, agent development, coding tools, and startup dynamics in 2026.
TechCrunch is promoting StrictlyVC San Francisco, a venture capital and founder conference taking place April 30 at Sentro Filipino Cultural Center.
Maggie Appleton argues that public learning through blogging and streaming creates perception of competence that leads to exclusive professional opportunities.
Astronomers are adopting GPUs to accelerate galaxy discovery, contributing to global GPU demand pressures.
Beehiiv released new creator tools including webinars and customizable paywalls to consolidate creator business operations.
OpenAI released documentation on configuring Codex settings for personalization and workflow customization.
OpenAI published a guide to ten practical workplace uses for Codex across task automation, deliverable creation, and workflow integration.
India's non-gaming app market is growing, led by streaming and AI services, though global platforms capture most revenue gains.
Enterprise leaders are recognizing that AI systems require well-designed data fabric architecture to provide business context and enable safe, coordinated autonomous decision-making.
MIT Technology Review publishes a roundup of ten topics the publication considers important in AI.
Andreas Påhlsson-Notini argues current AI agents replicate human flaws like lack of focus and constraint-handling rather than transcending them.
Simon Willison comments on efforts to inject synthetic pelican-bicycle images into AI training datasets as a form of data poisoning.
Simon Willison releases llm-openrouter 0.6 with a refresh command for updating available models.
Simon Willison used Claude Code to add beat content support to his blog-to-newsletter tool via a single agentic prompt.
PyCon US 2026 in Long Beach will feature new dedicated AI and Security tracks alongside traditional programming.