# forck.live — 25–31 May 2026

> AI lab announcements, news reports and podcasts.

forck.live follows what companies announce about their own products. Press may
corroborate an official story; it never leads and never appears as an item of its
own. A missing date stays missing. A title is the publisher's words — where a
company does not publish in English, an English rendering is listed separately
as `Title in English` and named as this site's translation, never as theirs.

This file is one issue of the weekly archive: every confirmed
first-party announcement of ISO week 2026-W22, 25–31 May 2026, newest
first. The window never moves, so this URL always names the same seven days —
but a post published inside them and detected later is added when it arrives,
so a closed week can still gain entries.

Every entry is one first-party announcement, and carries the same eight fields
in this order before any others:

    Company · Announced · Category · Coverage · Announcement · Group · Source · Record

`Source` is the company's own page the announcement was traced to and `Record`
is this site's page for it. Both are safe to cite.

`Coverage` is a number of outlets, or `not counted` — which means this record
reached no count of its own, never that nobody wrote about it.

`Category` is this tracker's own analysis, and is authoritative wherever it is
stated. `not stated` means either that nothing has analysed the post yet or
that the analysis placed it outside this taxonomy; neither is a verdict about
the announcement.

`Announcement` is whether the post is news of something the company did, rather
than a how-to, a customer story, a changelog listing or a staffing note.

`Group` is which of this record's groups the entry falls in, carried on the
entry itself so it holds where the groups are headings and where they are not.
One of: `models`, `covered`, `announcements`, `routine`.

Anything after `Record` is held detail that only some announcements carry.

- This week's page: https://forck.live/week/2026-W22
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### OpenClaw — Safer Than YOLO: Auto Mode for Exec Approvals

- Company: OpenClaw (openclaw.ai)
- Announced: 2026-05-31
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openclaw.ai/blog/safer-than-yolo-auto-mode-for-exec-approvals/
- Record: https://forck.live/items/17764-safer-than-yolo-auto-mode-for-exec-approvals
- Subject: OpenClaw

OpenClaw introduced an opt-in 'auto' mode for host exec approvals, which runs commands that match policy without prompts, sends low-risk misses to a reviewer model, and routes uncertain cases to a human. The reviewer model can be set separately from the main agent model, and approvals can be routed to Slack, Telegram, or iMessage. The default mode remains unchanged.

### Google DeepMind — 9 demos of Gemini Omni and Gemini 3.5 in action

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-05-29T17:30:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-3-5-videos/
- Record: https://forck.live/items/1308-9-demos-of-gemini-omni-and-gemini-3-5-in-action
- Subject: Gemini

Google DeepMind announced two new models, Gemini Omni and Gemini 3.5, at Google I/O 2026 and released 9 demonstration videos showcasing their capabilities.

### Harvey — Harvey FORUM: From Piloting AI to Operationalizing it

- Company: Harvey (harvey.ai)
- Announced: 2026-05-29T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-forum-from-piloting-ai-to-operationalizing-it
- Record: https://forck.live/items/17560-harvey-forum-from-piloting-ai-to-operationalizing-it
- Subject: Harvey

Harvey hosted its FORUM New York event, where legal leaders discussed the shift from AI experimentation to implementation. Key themes included the move from AI assistants to AI agents, the need for visibility into AI adoption, and the expanding role of General Counsels in leading AI transformation. Harvey introduced Command Center for deployment visibility and Contract Intelligence for in-house contract management.

### OpenAI — How Braintrust turns customer requests into code with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-29T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/braintrust
- Record: https://forck.live/items/101-how-braintrust-turns-customer-requests-into-code-with-codex
- Subject: GPT / ChatGPT / API

Braintrust engineers use Codex with GPT-5.5 to run experiments and code faster.

### OpenAI — Boston Children’s uses AI to unlock new diagnoses

- Company: OpenAI (openai.com)
- Announced: 2026-05-29T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/boston-childrens-hospital
- Record: https://forck.live/items/100-boston-children-s-uses-ai-to-unlock-new-diagnoses
- Subject: GPT / ChatGPT / API

Boston Children's Hospital uses OpenAI technology to improve patient care, reduce operational burden, and help diagnose more than 40 rare disease cases.

### Alibaba — Qwen-VLA: From Understanding the World to Acting in It

- Company: Alibaba (alibaba.com)
- Announced: 2026-05-29T09:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwenvla
- Record: https://forck.live/items/4608-qwen-vla-from-understanding-the-world-to-acting-in-it
- Subject: Qwen

Alibaba announces Qwen-VLA, a general-purpose Vision-Language-Action model built on the Qwen multimodal backbone, designed to extend visual perception, language understanding, and spatial reasoning into continuous action generation and trajectory prediction for embodied intelligence tasks such as robotic manipulation and vision-language navigation.

### Vercel — Protecting against token theft

- Company: Vercel (vercel.com)
- Announced: 2026-05-29T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/protecting-against-token-theft
- Record: https://forck.live/items/20707-protecting-against-token-theft
- Subject: v0 / AI SDK / AI Gateway

HTTP requests are inexpensive. Vercel charges ~$2/million, a fraction of a cent per call.

### OpenAI — Strengthening societal resilience with Rosalind Biodefense

- Company: OpenAI (openai.com)
- Announced: 2026-05-29T03:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/strengthening-societal-resilience-with-rosalind-biodefense
- Record: https://forck.live/items/102-strengthening-societal-resilience-with-rosalind-biodefense
- Subject: GPT / ChatGPT / API

OpenAI launches Rosalind Biodefense, expanding access to GPT-Rosalind for vetted developers and U.S. government partners in biodefense and public health.

### Krea — Krea 2 LoRA Training: Public LoRAs in Krea Train

- Company: Krea (krea.ai)
- Announced: 2026-05-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/public-loras-krea-2-train
- Record: https://forck.live/items/15075-krea-2-lora-training-public-loras-in-krea-train
- Subject: Krea 2 / AI creative platform

Krea launched a closed beta for Krea 2 LoRA training within its Train platform. Selected users can train lightweight custom models for Krea 2 Medium and Krea 2 Large, covering style, character, or object concepts, and choose to keep them private or publish them as public LoRAs for discovery and reuse.

### Cognition — Verifying Agentic Development at Scale

- Company: Cognition (cognition.com)
- Announced: 2026-05-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/testing-development
- Record: https://forck.live/items/7657-verifying-agentic-development-at-scale
- Subject: Devin / SWE

Cognition's Devin AI software engineer now autonomously verifies its work using computer use in the cloud, running tests in parallel. The company reports that for the first time, more Devin sessions are triggered asynchronously than interactively. To improve reliability, Devin writes test plans grounded in source code and uses deterministic scripts for common setup steps.

### xAI — Grok Build 0.1 on API

- Company: xAI (x.ai)
- Announced: 2026-05-29
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://x.ai/news/grok-build-0-1
- Record: https://forck.live/items/3461-grok-build-0-1-on-api
- Subject: Grok

Grok Build 0.1, described as xAI's fastest coding model, has been made available on the xAI API in public beta.

### Cursor — Auto-review Run Mode

- Company: Cursor (cursor.com)
- Announced: 2026-05-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/auto-review
- Record: https://forck.live/items/2603-auto-review-run-mode
- Subject: Cursor

Cursor introduces Auto-review, a new run mode that reduces approval prompts and enables safer execution for Shell, MCP, and Fetch tool calls, with allowlisted calls running immediately and sandboxable calls running in a sandbox, while other actions are handled by a classifier subagent.

### Together AI — How Together AI built the world’s fastest speech-to-text stack

- Company: Together AI (together.ai)
- Announced: 2026-05-29
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/how-together-ai-built-the-worlds-fastest-speech-to-text-stack
- Record: https://forck.live/items/2344-how-together-ai-built-the-world-s-fastest-speech-to-text-stack
- Subject: Inference platform

Together AI announced optimizations to their speech-to-text (ASR) stack that reduce latency and improve throughput. By using TensorRT with multi-profile engines for real audio shapes, moving the decoder loop entirely to the GPU with conditional CUDA graph nodes, and collapsing CPU process boundaries to eliminate redundant copies, they achieved the fastest ASR stack according to Artificial Analysis. The stack serves NVIDIA Parakeet-TDT 0.6B v3 and OpenAI Whisper Large v3, with Parakeet transcribing 20 hours of speech in under 10 seconds.

### Hugging Face — Profiling in PyTorch (Part 1): A Beginner's Guide to torch.profiler

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-29
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/torch-profiler
- Record: https://forck.live/items/1496-profiling-in-pytorch-part-1-a-beginner-s-guide-to-torch-profiler
- Subject: Platform

The torch.profiler module for profiling PyTorch code, with a focus on matrix multiplication and addition. It explains how to set up profiling, read profiler tables and traces, and understand CPU/GPU kernel interactions. No new AI product or model is announced.

### OpenAI — A shared playbook for trustworthy third party evaluations

- Company: OpenAI (openai.com)
- Announced: 2026-05-29
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/trustworthy-third-party-evaluations-foundations
- Record: https://forck.live/items/103-a-shared-playbook-for-trustworthy-third-party-evaluations
- Subject: GPT / ChatGPT / API

OpenAI publishes guidance on third-party AI evaluations for frontier systems, covering assessment of capabilities, safeguards, and validity.

### Google Research — A New Era of Discovery: Google Research at I/O 2026

- Company: Google Research (research.google)
- Announced: 2026-05-28T20:58:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://research.google/blog/a-new-era-of-innovation-google-research-at-io-2026/
- Record: https://forck.live/items/1338-a-new-era-of-discovery-google-research-at-i-o-2026
- Subject: Research

Google Research announces Gemini for Science, a suite of experimental tools for scientific discovery, including Computational Discovery, Hypothesis Generation, Literature Insights, and Science Skills. Also highlights publications of ERA and Co-Scientist in Nature, and the Paper Assistant Tool and Gemini Deep Think for accelerating research.

### Harvey — Opus 4.8, Now Live in Harvey

- Company: Harvey (harvey.ai)
- Announced: 2026-05-28T17:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/opus-4-8-now-live-in-harvey
- Record: https://forck.live/items/17561-opus-4-8-now-live-in-harvey
- Subject: Harvey

Harvey made Claude Opus 4.8 available in its platform. The model, Anthropic's latest frontier model, scored 10.4% on Harvey's Legal Agent Benchmark (up from 7.1%) and 91.1% on BigLaw Bench. It is rolling out to eligible clients in the US and EU.

### Black Forest Labs — FLUX VTO: Virtual Try-On at scale

- Company: Black Forest Labs (bfl.ai)
- Announced: 2026-05-28T16:37:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://bfl.ai/blog/flux-vto-virtual-try-on-at-catalog-scale
- Record: https://forck.live/items/3390-flux-vto-virtual-try-on-at-scale
- Subject: FLUX

Black Forest Labs announces FLUX VTO, a model for virtual try-on at scale.

### ElevenLabs — Introducing Dubbing v2

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-05-28T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/introducing-dubbing-v2
- Record: https://forck.live/items/3355-introducing-dubbing-v2
- Subject: Voice AI

ElevenLabs announced Dubbing v2, a new version of its dubbing product or service, based solely on the title of the announcement. No further details are provided in the source text.

### ElevenLabs — Voice AI for Greece

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-05-28T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/voice-ai-for-greece
- Record: https://forck.live/items/3354-voice-ai-for-greece
- Subject: Voice AI

Voice AI for Greece, but no further details are provided.

### OpenAI — How Endava builds an agentic organization with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-28T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/endava
- Record: https://forck.live/items/104-how-endava-builds-an-agentic-organization-with-codex
- Subject: GPT / ChatGPT / API

Endava uses Codex to build an agentic organization, accelerating software delivery and reducing requirements analysis from weeks to hours.

### OpenClaw — OpenClaw Is Getting Faster, Smaller, and Easier to Trust

- Company: OpenClaw (openclaw.ai)
- Announced: 2026-05-28
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openclaw.ai/blog/lighter-core-sharper-claws/
- Record: https://forck.live/items/17765-openclaw-is-getting-faster-smaller-and-easier-to-trust
- Subject: OpenClaw

OpenClaw has been getting faster and smaller at the same time. The performance work is visible in agent turns.

### Liquid AI — LFM2.5-8B-A1B: An Even Better On-Device Mixture of Experts

- Company: Liquid AI (liquid.ai)
- Announced: 2026-05-28
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.liquid.ai/blog/lfm2-5-8b-a1b
- Record: https://forck.live/items/16783-lfm2-5-8b-a1b-an-even-better-on-device-mixture-of-experts
- Subject: LFM / d1 models

Liquid AI released LFM2.5-8B-A1B, an update to its edge MoE model with a 128K context window (up from 32K), a doubled vocabulary (128K tokens) for better non-Latin script efficiency, and reasoning-only output with explicit chain-of-thought. The base and post-trained models are available on Hugging Face and the Liquid Playground.

### Krea — Moodboard Gallery in Krea 2 + Preview of Random and Auto

- Company: Krea (krea.ai)
- Announced: 2026-05-28
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/moodboard-gallery-krea-2-random-auto
- Record: https://forck.live/items/15076-moodboard-gallery-in-krea-2-preview-of-random-and-auto
- Subject: Krea 2 / AI creative platform

Krea has expanded the Moodboard Gallery in Krea 2 with thousands of new moodboards for faster style discovery. The update also previews two new selection modes, Random and Auto, which are designed to speed up early exploration by picking or automatically selecting moodboards based on context. Both modes are in preview and may evolve based on feedback.

### Amp — Plugins, Everywhere

- Company: Amp (ampcode.com)
- Announced: 2026-05-28
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/plugins-everywhere
- Record: https://forck.live/items/7742-plugins-everywhere
- Subject: Amp

Amp plugins can now display UI elements on the web, including notifications, confirmation dialogs, input fields, and select elements.

### Mistral AI — Introducing Search Toolkit

- Company: Mistral AI (mistral.ai)
- Announced: 2026-05-28
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/search-toolkit
- Record: https://forck.live/items/3294-introducing-search-toolkit
- Subject: Mistral / Le Chat

Mistral AI announces the Search Toolkit, a product for production search pipelines that can be used anywhere.

### Mistral AI — Vibe gets to work.

- Company: Mistral AI (mistral.ai)
- Announced: 2026-05-28
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/vibe-agent
- Record: https://forck.live/items/3293-vibe-gets-to-work
- Subject: Mistral / Le Chat

Mistral AI announces Vibe, a unified agent for long-horizon productivity and coding, launching with Work and Code modes, along with a new Vibe VS Code extension.

### Mistral AI — AI Now Summit 2026

- Company: Mistral AI (mistral.ai)
- Announced: 2026-05-28
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://mistral.ai/news/ai-now-summit-2026
- Record: https://forck.live/items/3292-ai-now-summit-2026
- Subject: Mistral / Le Chat

Mistral announces the AI Now Summit 2026, focusing on innovations for global enterprises.

### Cohere — What is Model Context Protocol? A practical guide to MCP

- Company: Cohere (cohere.com)
- Announced: 2026-05-28
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cohere.com/blog/guide-to-mcp
- Record: https://forck.live/items/3283-what-is-model-context-protocol-a-practical-guide-to-mcp
- Subject: Command

Cohere published a practical guide explaining the Model Context Protocol (MCP), a protocol for providing context to AI models.

### OpenAI — OpenAI’s Frontier Governance Framework

- Company: OpenAI (openai.com)
- Announced: 2026-05-28
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/openai-frontier-governance-framework
- Record: https://forck.live/items/105-openai-s-frontier-governance-framework
- Subject: GPT / ChatGPT / API

OpenAI announces its Frontier Governance Framework, describing how its AI safety, security, and risk practices align with emerging EU and California regulations.

### Google Research — Private analytics via zero-trust aggregation

- Company: Google Research (research.google)
- Announced: 2026-05-27T16:56:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/private-analytics-via-zero-trust-aggregation/
- Record: https://forck.live/items/1339-private-analytics-via-zero-trust-aggregation
- Subject: Research

Google Research introduces a private analytics solution that combines a new cryptographic aggregation protocol with trusted execution environments (TEEs) to achieve zero-trust security, enabling single-message submission and multi-layered defense.

### Sakana AI — DiffusionBlocks: Training Neural Networks One Block at a Time

- Company: Sakana AI (sakana.ai)
- Announced: 2026-05-27T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/diffusion-blocks
- Record: https://forck.live/items/4683-diffusionblocks-training-neural-networks-one-block-at-a-time
- Subject: Sakana models

Sakana AI introduces DiffusionBlocks, a method to train neural networks one block at a time by reinterpreting the forward pass as a diffusion process, reducing memory requirements while matching end-to-end performance across ViT, DiT, masked diffusion, autoregressive transformers, and recurrent-depth transformers. The paper was accepted at ICLR 2026.

### Vercel — How Conductor moved parallel coding agents from the laptop to the cloud with Vercel Sandbox

- Company: Vercel (vercel.com)
- Announced: 2026-05-27T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-conductor-moved-parallel-coding-agents-from-the-laptop-to-the-cloud-with-vercel-sandbox
- Record: https://forck.live/items/20708-how-conductor-moved-parallel-coding-agents-from-the-laptop-to-the-cloud-with
- Subject: v0 / AI SDK / AI Gateway

Conductor, a multi-agent IDE, moved its parallel coding agents from local machines to the cloud using Vercel Sandboxes. The new Cloud Workspaces layer lets developers run multiple agents remotely without hardware constraints, and the interface remains unchanged. Conductor is model-agnostic and works with Claude Code, Codex, and other coding agents.

### Harvey — Harvey’s Help, Now Just an Email Away

- Company: Harvey (harvey.ai)
- Announced: 2026-05-27T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/using-email-harvey
- Record: https://forck.live/items/17562-harvey-s-help-now-just-an-email-away
- Subject: Harvey

Harvey introduced Email Harvey, an early-access feature that lets lawyers interact with the platform via email to ask questions, generate files, and advance projects without switching applications. The feature supports tasks such as analyzing regulatory updates and revising strategy documents, with results delivered as attachments in a private email thread. Users can continue deeper work in the Harvey web app, where email threads are visible and findings can be saved to matter-specific vaults.

### Lovable — Introducing subagents: Lovable is now better at multitasking

- Company: Lovable (lovable.dev)
- Announced: 2026-05-27T12:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/subagents-in-lovable
- Record: https://forck.live/items/12079-introducing-subagents-lovable-is-now-better-at-multitasking
- Subject: Lovable / AI app builder

Lovable introduced subagents that the main agent can spawn to research, explore, and search a project in parallel, with each subagent receiving its own context window and read-only access. This speeds up builds on large projects and can lower costs by routing lighter work to cheaper models, while the main agent's context stays clean. Subagents are enabled automatically with no change to how users interact with Lovable.

### OpenAI — Cisco and OpenAI redefine enterprise engineering with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-27T11:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/cisco
- Record: https://forck.live/items/106-cisco-and-openai-redefine-enterprise-engineering-with-codex
- Subject: GPT / ChatGPT / API

Cisco and OpenAI are using Codex to help Cisco scale AI-native development, accelerate AI Defense work, and automate defect remediation.

### Runway — Introducing Runway MCP

- Company: Runway (runway.com)
- Announced: 2026-05-27T10:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/company-news/mcp
- Record: https://forck.live/items/7877-introducing-runway-mcp
- Subject: Gen / Aleph

Runway launched Runway MCP, an MCP server that connects Runway to agents and coding tools, enabling users to generate images and videos directly from their agent without switching context. The server provides access to multiple state-of-the-art models including Gen-4.5, Seedance 2.0, GPT Image 2, Kling 3.0, and Nano Banana Pro.

### OpenAI — Building self-improving tax agents with Codex

- Company: OpenAI (openai.com)
- Announced: 2026-05-27T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/building-self-improving-tax-agents-with-codex
- Record: https://forck.live/items/107-building-self-improving-tax-agents-with-codex
- Subject: GPT / ChatGPT / API

OpenAI, Thrive, and Crete built a self-improving tax agent using Codex to automate filings, improve accuracy, and accelerate workflows.

### Cua — Inside Windows computer-use: synthetic cursors for background agents

- Company: Cua (cua.ai)
- Announced: 2026-05-27
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cua.ai/blog/inside-windows-computer-use
- Record: https://forck.live/items/16821-inside-windows-computer-use-synthetic-cursors-for-background-agents
- Subject: Cua Driver / Lume / Cua Bench / Cua Fleets

Cua Driver now supports Windows background computer-use for agents such as Claude Code, Codex, and Hermes. The driver provides window pixels, UIA/MSAA accessibility trees, and an action layer for clicks, typing, scrolling, values, and verification, using a synthetic cursor separate from the user's physical pointer. It routes between different backends (e.g., PrintWindow, Windows.Graphics.Capture, UI Automation, MSAA, PostMessage, SendInput) depending on the target app type and action.

### Krea — Krea 2 API: Style Transfer, Moodboards & Pricing

- Company: Krea (krea.ai)
- Announced: 2026-05-27
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/krea-2-api-launch
- Record: https://forck.live/items/15077-krea-2-api-style-transfer-moodboards-pricing
- Subject: Krea 2 / AI creative platform

Krea launched an API for its Krea 2 foundation image model, offering two model sizes (Medium and Large) with style transfer, moodboard support, and a tunable creativity parameter. The API is accessible through partners such as Fal, Comfy, Runware, and Nous Research.

### Sesame — Getting started

- Company: Sesame (sesame.com)
- Announced: 2026-05-27
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.sesame.com/blog/getting-started
- Record: https://forck.live/items/8045-getting-started
- Subject: Maya / CSM

Sesame offers four personal agents (Maya, Miles, Simone, Charlie) in a Mobile Preview with voice and text communication, web search, deep research, notes, reminders, and post-call summaries. A Research Preview is available on web with voice-only calls. The company plans an Android version of the Mobile Preview and Sesame eyewear hardware in 2027.

### Sesame — Voice your curiosity

- Company: Sesame (sesame.com)
- Announced: 2026-05-27
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sesame.com/blog/voice-your-curiosity
- Record: https://forck.live/items/8044-voice-your-curiosity
- Subject: Maya / CSM

Sesame released an iOS preview of its lifelike conversation partners, available in the App Store. The preview includes new characters, voice-first features, memory per agent, and incognito mode. The experience is free during the initial rollout.

### Hume — Disentangling Emotion from Voice: A Cross- Sampling Approach for Expressive Voice Data

- Company: Hume (hume.ai)
- Announced: 2026-05-27
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.hume.ai/blog/disentangling-emotion-from-voice
- Record: https://forck.live/items/8039-disentangling-emotion-from-voice-a-cross-sampling-approach-for-expressive
- Subject: EVI / Octave

Hume presents a cross-product sampling approach for disentangling emotion from voice in expressive voice data.

### Amp — Proof of Human

- Company: Amp (ampcode.com)
- Announced: 2026-05-27
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/proof-of-human
- Record: https://forck.live/items/7744-proof-of-human
- Subject: Amp

Amp added the ability to require active passkey authentication for certain operations, such as remote controlling a thread, as an extra authentication factor and proof-of-human for future features. Workspace admins can enforce this requirement for members, and some privileged admin operations now always require it.

### Amp — Drop the Neo

- Company: Amp (ampcode.com)
- Announced: 2026-05-27
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/drop-the-neo
- Record: https://forck.live/items/7743-drop-the-neo
- Subject: Amp

Amp Neo is now available to everyone and has been renamed to Amp. The --take-me-back flag has been removed, but users can still access the old version via a previous CLI version.

### Cognition — More Devins in More Places

- Company: Cognition (cognition.com)
- Announced: 2026-05-27
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/series-d
- Record: https://forck.live/items/7658-more-devins-in-more-places
- Subject: Devin / SWE

Cognition raised over $1B at a $26B valuation in a Series D round led by Lux Capital, General Catalyst, and 8VC. Enterprise usage grew more than 10x since the start of the year, and run-rate revenue reached $492M. The company also launched the SWE-1.6 model.

### xAI — Use Grok in Kilo Code

- Company: xAI (x.ai)
- Announced: 2026-05-27
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://x.ai/news/grok-kilocode
- Record: https://forck.live/items/3462-use-grok-in-kilo-code
- Subject: Grok

Grok is now available in the Kilo Code open-source agentic coding platform, accessible through SuperGrok or X Premium+ subscriptions.

### MiniMax — MiniMax Agent Team: Built for Long-Running Tasks and Continuous Evolution

- Company: MiniMax (minimax.io)
- Announced: 2026-05-27
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.minimax.io/blog/minimax-agent-team-long-running-1779893953
- Record: https://forck.live/items/3397-minimax-agent-team-built-for-long-running-tasks-and-continuous-evolution
- Subject: MiniMax models

MiniMax announces an overall upgrade of its Agent, now named Mavis (MiniMax as a Jarvis), an AI butler designed for long-running tasks and continuous evolution.

### Mistral AI — Physics AI research that’s shaping the industry.

- Company: Mistral AI (mistral.ai)
- Announced: 2026-05-27
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/physics-ai-research
- Record: https://forck.live/items/3296-physics-ai-research-that-s-shaping-the-industry
- Subject: Mistral / Le Chat

Mistral AI announces published breakthroughs in physics AI research that are pushing the state of the art.

### Mistral AI — Introducing physics AI at Mistral: the foundation for engineering acceleration.

- Company: Mistral AI (mistral.ai)
- Announced: 2026-05-27
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://mistral.ai/news/introducing-physics-ai-at-mistral
- Record: https://forck.live/items/3295-introducing-physics-ai-at-mistral-the-foundation-for-engineering-acceleration
- Subject: Mistral / Le Chat

Mistral AI announces a new class of AI models that predict the behavior of physical systems, aimed at powering engineering and hardware products.

### Hugging Face — Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-27
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/delta-weight-sync
- Record: https://forck.live/items/1498-shipping-a-trillion-parameters-with-a-hub-bucket-delta-weight-sync-in-trl
- Subject: Platform

Hugging Face announces a new delta weight sync feature in TRL that encodes only the changed elements of model weights between RL optimizer steps as a sparse safetensors file, uploads it to a Hugging Face Bucket, and tells vLLM to fetch it, reducing per-step payload from 1.2 GB to 20-35 MB for Qwen3-0.6B.

### Hugging Face — Reachy Mini goes fully local

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-27
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/local-reachy-mini-conversation
- Record: https://forck.live/items/1497-reachy-mini-goes-fully-local
- Subject: Platform

To run the Reachy Mini conversation app fully locally using the speech-to-speech library, with recommended models including Gemma 4, Silero VAD, Parakeet-TDT 0.6B v3 STT, and Qwen3-TTS.

### OpenAI — Election information and safeguards in 2026

- Company: OpenAI (openai.com)
- Announced: 2026-05-27
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/election-safeguards-2026
- Record: https://forck.live/items/109-election-information-and-safeguards-in-2026
- Subject: GPT / ChatGPT / API

OpenAI announces measures to help people access information, support cyber defenders, and increase AI transparency ahead of global elections.

### OpenAI — Warp’s big bet on building open source with GPT-5.5

- Company: OpenAI (openai.com)
- Announced: 2026-05-27
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/warp
- Record: https://forck.live/items/108-warp-s-big-bet-on-building-open-source-with-gpt-5-5
- Subject: GPT / ChatGPT / API

Warp uses GPT-5.5 and OpenAI models to coordinate coding agents across local, cloud, and open-source development workflows.

### ElevenLabs — Introducing Music v2

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-05-26T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/introducing-music-v2
- Record: https://forck.live/items/3356-introducing-music-v2
- Subject: Voice AI

ElevenLabs announced the introduction of Music v2.

### Supabase — Protecting your Supabase projects from npm supply chain attacks

- Company: Supabase (supabase.com)
- Announced: 2026-05-26
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/protecting-your-supabase-projects-from-npm-supply-chain-attacks
- Record: https://forck.live/items/21285-protecting-your-supabase-projects-from-npm-supply-chain-attacks
- Subject: Supabase

Supabase published a guide on defending against npm supply chain attacks, covering attack patterns like maintainer compromise, typosquatting, and build pipeline compromise, and recommending mitigations such as upgrading to pnpm 11, pinning dependency versions, committing lockfiles, and disabling install scripts.

### Kyutai — Kairos: Understanding Data Temporality Impact on LLM pre-training

- Company: Kyutai (kyutai.org)
- Announced: 2026-05-26
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://kyutai.org/blog/2026-05-26-kairos
- Record: https://forck.live/items/18341-kairos-understanding-data-temporality-impact-on-llm-pre-training
- Subject: Moshi / Pocket TTS

← Back to Blog Kairos: Understanding Data Temporality Impact on LLM pre-training 26 May 2026

### Gradium — Phonon update: 1.00% WER on Seed-TTS, smaller than every model we beat

- Company: Gradium (gradium.ai)
- Announced: 2026-05-26
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://gradium.ai/blog/phonon-update-may-2026
- Record: https://forck.live/items/18318-phonon-update-1-00-wer-on-seed-tts-smaller-than-every-model-we-beat
- Subject: Gradium TTS / Phonon

Gradium updated its 100M-parameter on-device TTS model Phonon, reducing word error rate from 1.48% to 1.00% on the Seed-TTS English benchmark and improving speaker similarity from 56.37% to 59.51%. The update removes the previous 100-token minimum input padding and adds int8 quantization support without audio quality loss. Phonon remains in private beta.

### Runway — Project Luxo: Crossing the Uncanny Valley of AI Media

- Company: Runway (runway.com)
- Announced: 2026-05-26
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://runway.com/news/company-news/project-luxo
- Record: https://forck.live/items/7878-project-luxo-crossing-the-uncanny-valley-of-ai-media
- Subject: Gen / Aleph

A harrowing story of loss set at sea as two young boys confront impossible conditions and constant danger is presented.

### LG AI Research — GPU Job Scheduling Using an Idle Inference GPU Pool

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-05-26
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=651
- Record: https://forck.live/items/4748-gpu-job-scheduling-using-an-idle-inference-gpu-pool
- Subject: EXAONE

LG AI Research share how they maximize GPU resource utilization by scheduling jobs on idle inference GPU pools during off-peak hours.

### MiniMax — Why Can't the MiniMax LLM Say "Ma Jiaqi"? Internal Investigation of Sparse Token Forgetting

- Company: MiniMax (minimax.io)
- Announced: 2026-05-26
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.minimax.io/blog/sparse-token-forgetting
- Record: https://forck.live/items/3398-why-can-t-the-minimax-llm-say-ma-jiaqi-internal-investigation-of-sparse-token
- Subject: MiniMax models

MiniMax published an internal investigation into the sparse token forgetting phenomenon affecting the MiniMax M2 series of LLMs.

### Krea — Animate Krea 2 illustrations with Seedance 2.0

- Company: Krea (krea.ai)
- Announced: 2026-05-25
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.krea.ai/blog/krea-2-seedance-animation
- Record: https://forck.live/items/15103-animate-krea-2-illustrations-with-seedance-2-0
- Subject: Krea 2 / AI creative platform

Krea published a guide showing how to animate illustrations created with Krea 2 using Seedance 2.0 within the same Krea workflow. The process involves generating a style asset in Krea 2, then selecting Seedance 2.0 in the video tool and using the Krea 2 image as the start frame. Users can add multiple Krea 2 generations and prompt duration and action for each shot individually.

### Amp — GPT Image 2 Paints Better

- Company: Amp (ampcode.com)
- Announced: 2026-05-25
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/gpt-image-2
- Record: https://forck.live/items/7745-gpt-image-2-paints-better
- Subject: Amp

Amp's painter tool now uses GPT Image 2, which is a better image editor than Gemini 3 Pro Image at preserving text, typography, and visual style when editing UI screenshots, at roughly one-quarter the price.

### xAI — Introducing Grok Build

- Company: xAI (x.ai)
- Announced: 2026-05-25
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/grok-build-cli
- Record: https://forck.live/items/3463-introducing-grok-build
- Subject: Grok

xAI announces Grok Build, a new coding agent available in early beta for SuperGrok and X Premium Plus subscribers, running from the terminal.

### Hugging Face — Harness, Scaffold, and the AI Agent Terms Worth Getting Right

- Company: Hugging Face (huggingface.co)
- Announced: 2026-05-25
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/agent-glossary
- Record: https://forck.live/items/1499-harness-scaffold-and-the-ai-agent-terms-worth-getting-right
- Subject: Platform

A glossary post clarifying the definitions of AI agent terms such as model, scaffolding, harness, and agent, distinguishing between concepts that are often confused.

### OpenAI — OpenAI, Grupo Folha and Grupo UOL announce strategic content partnership

- Company: OpenAI (openai.com)
- Announced: 2026-05-25
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/grupo-folha-grupo-uol-partnership
- Record: https://forck.live/items/111-openai-grupo-folha-and-grupo-uol-announce-strategic-content-partnership
- Subject: GPT / ChatGPT / API

OpenAI partners with Grupo Folha and Grupo UOL to bring Brazilian journalism to ChatGPT with attribution and transparency.

### OpenAI — AI is becoming a first hire for small businesses

- Company: OpenAI (openai.com)
- Announced: 2026-05-25
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/ai-first-hire-small-business
- Record: https://forck.live/items/110-ai-is-becoming-a-first-hire-for-small-businesses
- Subject: GPT / ChatGPT / API

Research indicates that 4 million Americans use ChatGPT to start, run, and grow small businesses, reducing the cost of entrepreneurship with AI assistance.