AI lab announcements, 27 April – 3 May 2026

35 announcements from 13 labs, first-party only

OpenAI

OpenAICapability change· 30 Apr

Introducing Advanced Account Security

OpenAI announced Advanced Account Security, which includes phishing-resistant login, stronger recovery, and enhanced protections to safeguard sensitive data and prevent account takeover.

OpenAI· 29 Apr

Where the goblins came from

The article discusses the spread of 'goblin outputs' in AI models, specifically focusing on GPT-5, and covers the timeline, root cause, and fixes for these personality-driven quirks.

OpenAIInfrastructure· 29 Apr

Building the compute infrastructure for the Intelligence Age

OpenAI is expanding its Stargate project to build more data center capacity for AGI compute infrastructure.

OpenAI· 29 Apr

Cybersecurity in the Intelligence Age

OpenAI announces a five-part action plan to strengthen cybersecurity in the Intelligence Age, with an emphasis on democratizing AI-powered cyber defense and protecting critical systems.

OpenAISafety or policy· 28 Apr

Our commitment to community safety

OpenAI describes how it protects community safety within ChatGPT via model safeguards, misuse detection, policy enforcement, and collaboration with safety experts.

OpenAIAvailability change· 28 Apr

OpenAI models, Codex, and Managed Agents come to AWS

OpenAI announces availability of GPT models, Codex, and Managed Agents on AWS.

OpenAIAvailability change· 27 Apr

OpenAI available at FedRAMP Moderate

OpenAI has achieved FedRAMP Moderate authorization, making ChatGPT Enterprise and the OpenAI API available to U.S. federal agencies.

OpenAIPartnership· 27 Apr

The next phase of the Microsoft OpenAI partnership

OpenAI and Microsoft announce an amended agreement that simplifies their partnership, adds long-term clarity, and supports continued AI innovation at scale.

OpenAI· 27 Apr

Choco automates food distribution with AI agents

This is a customer story about how Choco used OpenAI APIs to improve food distribution, productivity, and growth.

OpenAIOpen-weight release· 27 Apr

An open-source spec for orchestration: Symphony

OpenAI announces Symphony, an open-source specification for orchestrating Codex agents, designed to connect issue trackers to persistent agent systems.

IBM's Granite 4.1 family of dense, decoder-only LLMs (3B, 8B, 30B) is introduced, trained from scratch on ~15T tokens with a multi-stage pre-training pipeline including long-context extension to 512K…

Hugging FacePartnership· 29 Apr

DeepInfra on Hugging Face Inference Providers 🔥

DeepInfra is now a supported Inference Provider on Hugging Face Hub, enabling access to models like DeepSeek V4, Kimi-K2.6, GLM-5.1, and DeepSeek V4 Pro through serverless inference.

Hugging FaceNew model· 28 Apr

Introducing NVIDIA Nemotron 3 Nano Omni: Long-Context Multimodal Intelligence for Documents, Audio and Video Agents

NVIDIA announces Nemotron 3 Nano Omni, a new omni-modal understanding model that extends the Nemotron multimodal line to handle text, image, video, and audio inputs.

Hugging FaceStory of the weekDeveloper tool· 27 Apr

How to build scalable web apps with OpenAI's Privacy Filter

OpenAI released Privacy Filter, an open-source PII detector, on Hugging Face Hub. The blog post demonstrates building three web apps (Document Privacy Explorer, Image Anonymizer, SmartRedact Paste)…

Together AI

Together AIPartnership· 30 Apr

Announcing Together AI and Adaption Partnership

Together AI announces a partnership with Adaption to integrate Together Fine-Tuning into Adaptive Data platform, enabling users to fine-tune models on optimized datasets directly from the platform.

Together AI· 30 Apr

From 732 bytes to nowhere: shutting down Copy Fail in production

Together AI responded to the Copy Fail kernel vulnerability (CVE-2026-31431) by disabling the vulnerable algif_aead module across its infrastructure, unloading it from running kernels and removing…

Together AIAvailability change· 29 Apr

DeepSeek-V4 Pro now available on Together AI

DeepSeek V4 Pro is now available on Together AI with a 512K-token context window, using a 1.6T-parameter MoE architecture with 49B activated parameters, supporting three reasoning modes (Non-Think,…

Together AIAvailability change· 28 Apr

Together AI Brings NVIDIA Nemotron 3 Nano Omni to Developers on Day 0

Together AI announces that NVIDIA Nemotron 3 Nano Omni is now available on its platform, offering developers immediate access to the multimodal model for production-scale inference.

Cursor

CursorCapability change· 1 May

Team Marketplace Updates

Admins can now create a team marketplace without needing to connect a repository first. The update also lets them add, remove, and configure install behavior for first-party plugins directly in the…

CursorCapability change· 30 Apr

Cursor Security Review

enters beta on Teams and Enterprise plans, introducing Security Reviewer for PR-level vulnerability checks and Vulnerability Scanner for scheduled codebase scans, with support for custom triggers,…

CursorDeveloper tool· 29 Apr

Build programmatic agents with the Cursor SDK

Cursor released a software development kit (SDK) for building programmatic agents, along with updates to its Cloud Agents API, available in public beta with token-based pricing.

Google DeepMind

Google DeepMindResearch paper· 30 Apr

Enabling a new model for healthcare with AI co-clinician

Google DeepMind is researching the development of an AI co-clinician for healthcare.

Google DeepMind· 27 Apr

Join the new AI Agents Vibe Coding Course from Google and Kaggle

Google announces the return of its 5-Day AI Agents Intensive Course in partnership with Kaggle, with registration now open.

Google DeepMindPartnership· 27 Apr

Announcing our partnership with the Republic of Korea

Google DeepMind announced a partnership with the Republic of Korea to accelerate scientific breakthroughs using frontier AI models.

Alibaba

AlibabaDeveloper tool· 30 Apr

Qwen-Scope: Decoding Intelligence, Unleashing Potential

Alibaba introduces Qwen-Scope, an interpretability toolkit that uses Sparse Autoencoders (SAEs) trained on Qwen3 and Qwen3.5 series models to decompose hidden representations into interpretable…

AlibabaDeveloper tool· 28 Apr

FlashQLA: CP-/Bwd-Friendly Fused Linear Attention Kernels for GDN

Alibaba's Qwen team open-sourced FlashQLA, a high-performance linear attention kernel library for GDN (Gated Delta Network) that achieves 2-3x forward and 2x backward speedup over the FLA Triton…

ElevenLabs

ElevenLabsProduct launch· 29 Apr

Introducing ElevenMusic

ElevenLabs announces ElevenMusic, a new product.

ElevenLabs· 28 Apr

ElevenLabs is expanding in Spain

Google Research

Google Research· 1 May

Catalyzing scientific impact through global partnerships and open resources

Google Research publishes a blog post describing its open science philosophy, partnerships with scientific organizations, and a catalog of open-source tools and datasets across genomics,…

Google ResearchCapability change· 29 Apr

Four ways Google Research scientists have been using Empirical Research Assistance

Google Research scientists have been using Empirical Research Assistance (ERA) for real-world applications in epidemiology, cosmology, atmospheric monitoring, and neuroscience, including generating…

Baidu

BaiduNew model· 30 Apr

ERNIE-5.1-Preview Tops LMArena Text Leaderboard as No.1 Chinese Model!

Baidu announced that ERNIE-5.1-Preview topped the LMArena Text leaderboard as the No. 1 Chinese model and achieved high rankings in several category leaderboards.

Mistral AI

Mistral AIAvailability change· 27 Apr

Workflows for work that runs the business

Mistral AI announced that Workflows is now in public preview.

Sakana AI

Sakana AIResearch paper· 28 Apr

KAME: Tandem Architecture for Enhancing Knowledge in Real-Time Speech-to-Speech Conversational AI

Sakana AI introduces KAME, a tandem architecture for real-time speech-to-speech conversational AI that uses a fast speech model and an asynchronous backend LLM to enable 'speak while thinking'.

Tencent

TencentResearch paper· 30 Apr

Real life is where context gets hard

Tencent Hunyuan introduces CL-bench Life, a benchmark for evaluating context learning ability in real-life settings, containing 405 context-task pairs and 5,348 human-written rubrics.

xAI

xAICapability change· 30 Apr

Custom Voices

xAI announced Custom Voices, a feature that allows users to clone a voice from a short recording and use it across Grok Text to Speech and Voice Agent APIs.