Researchers demonstrate that voice AI systems are vulnerable to hidden audio attacks (adversarial examples). These inaudible attacks can fool models and compromise the security of voice assistants.
OpenAI and Dell partner to deploy Codex in hybrid and on-premise enterprise environments. The partnership enables organizations to securely launch AI coding agents across proprietary data and workflows.
AI startups generated $80 billion in revenue, but Anthropic and OpenAI capture 89% of it. Market concentration remains extreme among top players.
The UK Government Digital Service (GDS) published guidance on May 14th recommending public sector organisations keep open source as default, implicitly criticising the NHS's decision to close repositories following vulnerabilities found via Project Glasswing. GDS argues closure increases costs and reduces reusability.
Google DeepMind releases Gemini for Science, a suite of AI tools and experiments designed to accelerate scientific research by expanding the scale and precision of scientific exploration.
Busy month with multiple flagship releases: Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, GLM-5.1. Nathan Lambert also covers CAISI's V4 assessment of these open-source models.
A Stanford geneticist uses Google DeepMind's Co-Scientist to identify existing medicines that could treat liver fibrosis. The AI tool helps discover candidate molecules from approved treatments.
Google DeepMind's WeatherNext AI model improved forecasts for Hurricane Melissa in Jamaica. The model gave National Hurricane Center forecasters additional lead time to warn communities ahead of the historic landfall.
Simon Willison releases inaturalist-clumper 0.1, an open-source tool to cluster and publish iNaturalist sightings on his blog. Running in production for several weeks.
Vercel adds OpenTelemetry trace generation via CLI. Commands `vercel curl --trace` and `vercel traces get` enable trace generation and retrieval by request ID. Available on all plans.
Release of datasette-agent 0.1a1. Version now uses `execute-sql` permission when deciding which tables to list to the user.
OpenAI integrates Codex into ChatGPT mobile app, enabling real-time monitoring, steering, and approval of coding tasks across devices and remote environments.
Hugging Face introduces an asynchronicity technique for optimizing continuous batching in inference servers. The method improves throughput by handling requests non-blockingly, reducing latency and increasing GPU resource utilization.
Vercel adds Firewall management via CLI. New `vercel firewall` command enables configuration of custom rules, IP blocks, system bypasses, and attack modes from the terminal. An agent skill documents safe rollout best practices.
Vercel Sandbox now supports Node.js 26. Users must upgrade @vercel/sandbox to 1.10.22 or 0.0-beta.19 (v2) and set runtime property to node26.
GitLab announces workforce reduction and restructuring for the agentic era: cutting 30% of countries with small teams (out of ~60), flattening organization (removing 3 management layers), and reorganizing R&D into ~60 smaller autonomous teams with end-to-end ownership.
Simon Willison documents using LLM in script shebang lines. The LLM CLI supports fragments (-f), tool calls (-T), and YAML templates defining Python functions. Examples: generate SVG, write haiku with current time, or run calculations with gpt-5.4-mini.
Tobias Lütke describes River, Shopify's internal coding agent tool, operating entirely in public on Slack. River declines direct messages and enforces public channel conversations, creating an osmosis learning environment where all employees observe work and learn from each other without formal curriculum.
ChatGPT saw rapid adoption growth in Q1 2026, fastest among users over 35 with more balanced gender distribution. Data signals mainstream AI adoption beyond early adopters.
Thariq Shihipar (Claude Code team, Anthropic) advocates HTML over Markdown for Claude outputs. HTML enables SVG diagrams, interactive widgets, and in-page navigation, improving clarity for complex explanations (e.g., PR reviews with inline annotations).
Hugging Face introduces EMO, a pretrained mixture of experts (MoE) model designed to develop emergent modularity. The approach aims to create specialized experts that naturally form during training, improving model efficiency and performance.
Vercel AI SDK adds web adapter support to build browser-based chat UIs. Developers can define bots on the server and stream replies to the browser using the React useChat hook.
llm-gemini 0.31 released with support for gemini-3.1-flash-lite, now generally available after preview status. Model unchanged since March preview.
Parloa leverages OpenAI models to build scalable voice-driven customer service agents. The platform enables enterprises to design, simulate, and deploy reliable real-time customer interactions.
Uber deploys OpenAI models to power AI assistants and voice features for drivers (earnings optimization) and riders (faster booking) across its global real-time marketplace platform.
Singular Bank built Singularity, an internal assistant using ChatGPT and Codex, enabling bankers to save 60–90 minutes daily on meeting prep, portfolio analysis, and follow-ups.
Hugging Face adds anti-Benchmaxxer filtering to the open ASR leaderboard to prevent artificial benchmark optimization. The system detects models over-optimized for test metrics without real generalization.
OpenAI releases B2B Signals research showing how frontier enterprises deepen AI adoption, scale Codex-powered agentic workflows, and build durable competitive advantage.
OpenAI releases System Card for GPT-5.5 Instant, documenting model capabilities and limitations. Official technical documentation with no availability or pricing details announced.
Vercel adds a CLI command to query observability metrics. Teams with Observability Plus can access performance, reliability, and security data. Coding agents can leverage this feature to debug and analyze applications.
KIKO Milano migrated its e-commerce infrastructure from AWS EC2 to Vercel. Results: eliminated 3 weeks of Black Friday prep, 75% reduction in build times, shifted from minimal to multiple daily deployments. Team replaced manual scaling with focus on customer experience.
OpenAI scales Stargate infrastructure to support AGI development, adding new data center capacity to meet surging AI computational demand.
Vercel integrates Pro plan management directly into Stripe Projects via CLI. Developers can provision or upgrade subscriptions without leaving the terminal, using secure shared payment tokens.
Vercel held its 2026 AI Accelerator with 39 teams pitching on April 16th. Over 6 weeks, founders built agents, developer tools, consumer apps, and vertical AI for finance, security, healthcare, and robotics. Each team received 200K+ in credits. Top three teams won 100K+ in prizes and credits, with first place receiving investment from Vercel Ventures.
Starting April 29th, Vercel Hobby projects will have a maximum 30-day deployment retention. Expired deployments will be automatically deleted, except the 10 most recent and aliased ones. Pro and Enterprise plans unaffected.
Choco deployed OpenAI APIs to automate food distribution using AI agents, improving productivity and enabling growth. In-depth customer case study demonstrating real-world operational AI impact in supply chain.
OpenAI releases Privacy Filter, an open-weight model for detecting and redacting personally identifiable information (PII) in text with state-of-the-art accuracy. No specific performance metrics or benchmarks disclosed in the announcement.
Hyatt rolls out ChatGPT Enterprise globally across its workforce, leveraging GPT-5.4 and Codex to boost productivity, operations, and guest experience.
CRUX is a new evaluation project for measuring frontier AI capabilities on long, messy open-world tasks, moving beyond traditional benchmarks.
Cloudflare integrates OpenAI's GPT-5.4 and Codex into Agent Cloud, enabling enterprises to build, deploy, and scale AI agents for production tasks. Platform emphasizes agentic workflows with security and speed.