# forck.live — 26 January – 1 February 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-W05, 26 January – 1 February 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-W05
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Sarvam AI — Sarvam Dub: State-of-the-Art Dubbing for Indian Languages

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-02-01T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/sarvam-dub/
- Record: https://forck.live/items/18399-sarvam-dub-state-of-the-art-dubbing-for-indian-languages
- Subject: Sarvam / Bulbul / Saaras

AI has changed the way content moves. For creators and content teams, this shift raises the bar.

### OpenAI — Silver lining playbook: Likely China-origin activity targeting US persons

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-silver-lining-playbook
- Record: https://forck.live/items/11691-silver-lining-playbook-likely-china-origin-activity-targeting-us-persons
- Subject: GPT / ChatGPT / API

OpenAI banned a small set of ChatGPT accounts likely originating in mainland China that used the platform to research US persons, federal building locations, and social-engineering tactics. The accounts generated English-language email drafts impersonating a Hong Kong consulting firm to target US state-level officials and policy analysts with recruitment invitations, and requested guidance on face-swapping software. OpenAI found no evidence the accounts successfully elicited responses from their targets.

### OpenAI — Romance scams: AI-enabled romance scam workflows

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-romance-scam
- Record: https://forck.live/items/11690-romance-scams-ai-enabled-romance-scam-workflows
- Subject: GPT / ChatGPT / API

OpenAI banned accounts that used its AI to support romance scam workflows, including outreach, translation, victim engagement, and investment-fraud lures. The company describes a common pattern in these scams: the ping (cold contact), the zing (generate emotion), and the sting (extract money). This case study is part of OpenAI's February 2026 report on disrupting malicious uses of AI.

### OpenAI — Operation “False Witness”: Fake recovery service impersonating authorities

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-false-witness
- Record: https://forck.live/items/11689-operation-false-witness-fake-recovery-service-impersonating-authorities
- Subject: GPT / ChatGPT / API

OpenAI banned a cluster of ChatGPT accounts that very likely originated in Cambodia and used AI to pose as recovery services, law firms, and authorities in a scam targeting fraud victims. The scammers used ChatGPT to generate promotional content, impersonate attorneys and the FBI's IC3, and draft messages requesting advance fees and cryptocurrency payments. OpenAI stated that its policies prohibit using its tools for fraud and that it is collaborating with industry peers and authorities to disrupt such activity.

### OpenAI — Operation “Date Bait”: AI-enabled scam targeting loveseekers

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-date-bait
- Record: https://forck.live/items/11688-operation-date-bait-ai-enabled-scam-targeting-loveseekers
- Subject: GPT / ChatGPT / API

OpenAI banned a cluster of ChatGPT accounts and one API customer that very likely originated in Cambodia and used AI to run a semi-automated romance and task scam targeting Indonesian loveseekers. The operation used ChatGPT to generate promotional texts, pose as a flirtatious receptionist, translate messages, and create fake dating platforms. OpenAI states its policies strictly prohibit use of its tools for fraud or scams.

### OpenAI — Operation “No Bell”: Coordinated criticism of the US and allies

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-no-bell
- Record: https://forck.live/items/11687-operation-no-bell-coordinated-criticism-of-the-us-and-allies
- Subject: GPT / ChatGPT / API

OpenAI banned a ChatGPT account linked to a previously unreported influence operation, dubbed "No Bell", that used AI to generate long-form articles and social media content criticizing the US and its allies for audiences across Africa. The account likely originated in Russia and produced content praising Russia while criticizing Ukraine, the UK, and the US. The operation involved publishing articles under a fake byline and using techniques to conceal AI generation.

### OpenAI — Operation “Fish Food”: Russia-origin content farm activity

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-fish-food
- Record: https://forck.live/items/11686-operation-fish-food-russia-origin-content-farm-activity
- Subject: GPT / ChatGPT / API

OpenAI banned ChatGPT accounts linked to the Rybar network, which used AI to generate content for social media and websites across multiple languages, including Russian, English, and Spanish. The operation also involved drafting commercial plans for covert interference campaigns in Africa, with estimated budgets up to $600,000. The content typically praised Russia and its allies while criticizing Ukraine and Western countries.

### OpenAI — “Cyber Special Operations”: China-linked influence planning

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-cyber-special-operations
- Record: https://forck.live/items/11685-cyber-special-operations-china-linked-influence-planning
- Subject: GPT / ChatGPT / API

OpenAI banned a ChatGPT account linked to an individual associated with Chinese law enforcement who used the model to plan a covert influence operation targeting Japanese Prime Minister Sanae Takaichi and to polish status reports on 'cyber special operations.' The model refused to assist with the planning, and the user's activity revealed a large-scale, resource-intensive strategy by Chinese law enforcement to suppress dissent and silence critics, including the use of locally deployed AI models such as DeepSeek and Qwen.

### OpenAI — Operation “Trolling Stone”: Russia-linked influence activity

- Company: OpenAI (openai.com)
- Announced: 2026-02-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-trolling-stone
- Record: https://forck.live/items/11670-operation-trolling-stone-russia-linked-influence-activity
- Subject: GPT / ChatGPT / API

OpenAI banned accounts linked to a previously unreported influence operation called "Trolling Stone," which used AI to generate comments about an alleged Russian cult leader's arrest in Argentina.

### Vercel — Run untrusted code with Vercel Sandbox, now generally available

- Company: Vercel (vercel.com)
- Announced: 2026-01-30T13:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/vercel-sandbox-is-now-generally-available
- Record: https://forck.live/items/20757-run-untrusted-code-with-vercel-sandbox-now-generally-available
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the general availability of Vercel Sandbox, an execution layer for AI agents that provides on-demand, isolated Linux microVMs. The Sandbox is built on Vercel's own compute platform, Hive, and is designed for sub-second starts, full isolation, and ephemeral operation. Vercel is also open-sourcing the Sandbox CLI and SDK.

### Google DeepMind — Project Genie: Experimenting with infinite, interactive worlds

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-01-29T17:01:05+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/project-genie-experimenting-with-infinite-interactive-worlds/
- Record: https://forck.live/items/1256-project-genie-experimenting-with-infinite-interactive-worlds
- Subject: Gemini

Google AI Ultra subscribers in the U.S. can try out Project Genie, an experimental research prototype for creating and exploring worlds.

### OpenAI — Inside OpenAI’s in-house data agent

- Company: OpenAI (openai.com)
- Announced: 2026-01-29T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/inside-our-in-house-data-agent
- Record: https://forck.live/items/287-inside-openai-s-in-house-data-agent
- Subject: GPT / ChatGPT / API

OpenAI describes building an in-house AI data agent that uses GPT-5, Codex, and memory to reason over large datasets and provide insights quickly.

### OpenClaw — Introducing OpenClaw

- Company: OpenClaw (openclaw.ai)
- Announced: 2026-01-29
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openclaw.ai/blog/introducing-openclaw/
- Record: https://forck.live/items/17771-introducing-openclaw
- Subject: OpenClaw

The open-source agent platform formerly known as Clawd and Moltbot has been renamed OpenClaw. The rebrand includes new Twitch and Google Chat channel plugins, support for KIMI K2.5 and Xiaomi MiMo-V2-Flash models, image sending in web chat, and 34 security commits. The project remains an open agent platform that runs on user-controlled infrastructure.

### Amp — Slashing Custom Commands

- Company: Amp (ampcode.com)
- Announced: 2026-01-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/slashing-custom-commands
- Record: https://forck.live/items/7759-slashing-custom-commands
- Subject: Amp

Amp is removing custom commands and migrating them to skills, as skills now support direct user invocation. The post provides automated and manual migration steps for users.

### Baidu — PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing

- Company: Baidu (baidu.com)
- Announced: 2026-01-29
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://ernie.baidu.com/blog/posts/paddleocr-vl-1.5/
- Record: https://forck.live/items/4644-paddleocr-vl-1-5-towards-a-multi-task-0-9b-vlm-for-robust-in-the-wild-document
- Subject: ERNIE

Baidu released PaddleOCR-VL-1.5, an upgraded 0.9B vision-language model for document parsing, achieving 94.5% accuracy on OmniDocBench v1.5 and adding seal recognition and text spotting capabilities.

### Hugging Face — Introducing Daggr: Chain apps programmatically, inspect visually

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-29
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/daggr
- Record: https://forck.live/items/1555-introducing-daggr-chain-apps-programmatically-inspect-visually
- Subject: Platform

Hugging Face introduces Daggr, an open-source Python library for building AI workflows that connect Gradio apps, ML models, and custom functions, with automatic visual canvas generation for inspecting and rerunning steps.

### OpenAI — Taisei Corporation shapes the next generation of talent with AI

- Company: OpenAI (openai.com)
- Announced: 2026-01-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/taisei
- Record: https://forck.live/items/289-taisei-corporation-shapes-the-next-generation-of-talent-with-ai
- Subject: GPT / ChatGPT / API

Taisei Corporation's HR team is deploying ChatGPT Enterprise to advance AI-powered talent development.

### OpenAI — Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2026-01-29
- Category: deprecation
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/retiring-gpt-4o-and-older-models
- Record: https://forck.live/items/288-retiring-gpt-4o-gpt-4-1-gpt-4-1-mini-and-openai-o4-mini-in-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI will retire GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini from ChatGPT on February 13, 2026.

### Alibaba — Qwen3-ASR & Qwen3-ForcedAligner is Now Open Sourced: Robust, Streaming and Multilingual!

- Company: Alibaba (alibaba.com)
- Announced: 2026-01-28T16:00:04+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen3asr
- Record: https://forck.live/items/4597-qwen3-asr-qwen3-forcedaligner-is-now-open-sourced-robust-streaming-and
- Subject: Qwen

Alibaba's Qwen team open-sourced the Qwen3-ASR family: two ASR models (1.7B and 0.6B parameters) and a forced alignment model (0.6B), all under Apache 2.0 license.

### Lovable — Build more. Manage less.

- Company: Lovable (lovable.dev)
- Announced: 2026-01-28T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/a-smarter-lovable
- Record: https://forck.live/items/12110-build-more-manage-less
- Subject: Lovable / AI app builder

Lovable now plans features before coding, queues multiple prompts, and runs automated end-to-end tests. The company reports a 71% improvement in solving complex tasks. Google sign-in is available for Lovable Cloud apps.

### Vercel — How Stripe built a game-changing app in a single flight with v0

- Company: Vercel (vercel.com)
- Announced: 2026-01-28T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-stripe-built-a-game-changing-app-in-a-single-flight-with-v0
- Record: https://forck.live/items/20758-how-stripe-built-a-game-changing-app-in-a-single-flight-with-v0
- Subject: v0 / AI SDK / AI Gateway

Stripe's Business Value Consulting team used Vercel's v0 to build a customer-facing value calculator. The tool, prototyped by a non-engineer during a single flight, replaced spreadsheet-based financial models and led to a 288% increase in weekly unique users and an estimated 30-40 hours of saved FTE time per week.

### Harvey — Harvey In Practice: How Litigation Teams Manage Cases From Intake to Appeal

- Company: Harvey (harvey.ai)
- Announced: 2026-01-28T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-in-practice-how-litigation-teams-use-harvey
- Record: https://forck.live/items/17621-harvey-in-practice-how-litigation-teams-manage-cases-from-intake-to-appeal
- Subject: Harvey

Harvey published a guide describing how litigation teams use its AI platform across case stages—initial assessment, discovery, briefing, trial preparation, and appeal—to manage records, draft documents, and prepare for arguments. The post includes examples from law firms Mijares and Lynn Pinker Hurst & Schwegmann, where Harvey reduced review time from hours to minutes.

### ElevenLabs — Revolut selects ElevenLabs Agents to bolster customer support

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-01-28T12:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://elevenlabs.io/blog/revolut
- Record: https://forck.live/items/3366-revolut-selects-elevenlabs-agents-to-bolster-customer-support
- Subject: Voice AI

ElevenLabs announces that Revolut has chosen ElevenLabs Agents to enhance its customer support.

### Google Research — Towards a science of scaling agent systems: When and why agent systems work

- Company: Google Research (research.google)
- Announced: 2026-01-28T11:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/towards-a-science-of-scaling-agent-systems-when-and-why-agent-systems-work/
- Record: https://forck.live/items/1372-towards-a-science-of-scaling-agent-systems-when-and-why-agent-systems-work
- Subject: Research

Google Research publishes a paper introducing the first quantitative scaling principles for AI agent systems. The study evaluates 180 agent configurations across five architectures and finds that multi-agent coordination improves performance on parallelizable tasks but degrades it on sequential ones.

### OpenAI — EMEA Youth & Wellbeing Grant

- Company: OpenAI (openai.com)
- Announced: 2026-01-28T01:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/emea-youth-and-wellbeing-grant
- Record: https://forck.live/items/291-emea-youth-wellbeing-grant
- Subject: GPT / ChatGPT / API

OpenAI announces a €500,000 grant program for NGOs and researchers focusing on youth safety and wellbeing in the age of AI.

### OpenAI — The next chapter for AI in the EU

- Company: OpenAI (openai.com)
- Announced: 2026-01-28T01:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/the-next-chapter-for-ai-in-the-eu
- Record: https://forck.live/items/290-the-next-chapter-for-ai-in-the-eu
- Subject: GPT / ChatGPT / API

OpenAI is launching the EU Economic Blueprint 2.0, a set of initiatives including new data, partnerships, and efforts to boost AI adoption and skills across Europe.

### Gradium — Invincible Voice: How Gradium's Real-Time Voice AI Helps ALS Patients Speak Again

- Company: Gradium (gradium.ai)
- Announced: 2026-01-28
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://gradium.ai/blog/invincible-voice
- Record: https://forck.live/items/18330-invincible-voice-how-gradium-s-real-time-voice-ai-helps-als-patients-speak-again
- Subject: Gradium TTS / Phonon

Gradium announced a partnership powering Invincible Voice, an open-source assistive voice AI system led by Kyutai. Gradium provides the core voice AI API and models for the system, which combines speech recognition, natural language understanding, and voice synthesis to help people with ALS, stroke, or other conditions communicate in real time.

### Genspark — Genspark AI Workspace 2.0: Don''t Type, Just Speak, Work Gets Done

- Company: Genspark (genspark.ai)
- Announced: 2026-01-28
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/genspark-ai-workspace-2
- Record: https://forck.live/items/18261-genspark-ai-workspace-2-0-don-t-type-just-speak-work-gets-done
- Subject: Genspark

### Factory AI — Wipro and Factory Partner to Accelerate Agent-Native Software Development For Enterprises Globally

- Company: Factory AI (factory.com)
- Announced: 2026-01-28
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/wipro
- Record: https://forck.live/items/17877-wipro-and-factory-partner-to-accelerate-agent-native-software-development-for
- Subject: Droid

Wipro and Factory announced a strategic partnership to help enterprises adopt agent-native software development. Wipro will integrate Factory's platform into its WEGA delivery platform and roll it out across tens of thousands of engineers. Wipro Ventures also participated in Factory's recent funding round.

### Cua — A Story of Computer-Use: Where We Started, Where We're Headed

- Company: Cua (cua.ai)
- Announced: 2026-01-28
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cua.ai/blog/clawdbot-computer-use-history
- Record: https://forck.live/items/16824-a-story-of-computer-use-where-we-started-where-we-re-headed
- Subject: Cua Driver / Lume / Cua Bench / Cua Fleets

Cua co-founder Francesco Bonacci recounts the history of computer-use AI, from OpenAI's 2016 Universe platform to the viral open-source assistant Clawdbot (rebranded to Moltbot). The post traces key milestones including GPT-4V's vision capabilities, early browser agents, benchmarks like OSWorld and Windows Agent Arena, and the conceptual framing of LLMs as operating systems.

### Amp — Go Deep

- Company: Amp (ampcode.com)
- Announced: 2026-01-28
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/deep-mode
- Record: https://forck.live/items/7760-go-deep
- Subject: Amp

Amp introduced a new agent mode called 'deep' that uses GPT-5.2-Codex and a selection of tools. The mode is designed to silently read files and work autonomously for extended periods before making changes, complementing the existing 'smart' mode.

### Cognition — Devin crosses the pond: Cognition opens its London office

- Company: Cognition (cognition.com)
- Announced: 2026-01-28
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cognition.com/blog/cognition-london
- Record: https://forck.live/items/7682-devin-crosses-the-pond-cognition-opens-its-london-office
- Subject: Devin / SWE

Cognition is opening a London office to work more closely with European customers, particularly in the finance sector, where its autonomous software engineering product Devin has been adopted by large financial institutions.

### Cognition — Cognizant partners with Cognition to scale Devin and Windsurf across its engineering organization and global clients

- Company: Cognition (cognition.com)
- Announced: 2026-01-28
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/cognizant-cognition
- Record: https://forck.live/items/7681-cognizant-partners-with-cognition-to-scale-devin-and-windsurf-across-its
- Subject: Devin / SWE

Cognizant has partnered with Cognition to deploy Devin and Windsurf across its engineering organization and global client base. Cognition will embed a team of AI engineers to support the rollout.

### xAI — Grok Imagine API

- Company: xAI (x.ai)
- Announced: 2026-01-28
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/grok-imagine-api
- Record: https://forck.live/items/3475-grok-imagine-api
- Subject: Grok

xAI announces the Grok Imagine API, a state-of-the-art video generation API that balances quality, cost, and latency.

### Hugging Face — We Got Claude to Build CUDA Kernels and teach open models!

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-28
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/upskill
- Record: https://forck.live/items/1556-we-got-claude-to-build-cuda-kernels-and-teach-open-models
- Subject: Platform

Hugging Face announces upskill, a tool to generate agent skills using large models (like Claude Opus 4.5) and transfer them to smaller open models, demonstrated on CUDA kernel writing for diffusers models.

### OpenAI — Keeping your data safe when an AI agent clicks a link

- Company: OpenAI (openai.com)
- Announced: 2026-01-28
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/ai-agent-link-safety
- Record: https://forck.live/items/292-keeping-your-data-safe-when-an-ai-agent-clicks-a-link
- Subject: GPT / ChatGPT / API

OpenAI describes built-in safeguards to protect user data when AI agents open links, preventing URL-based data exfiltration and prompt injection.

### Google Research — ATLAS: Practical scaling laws for multilingual models

- Company: Google Research (research.google)
- Announced: 2026-01-27T18:58:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/atlas-practical-scaling-laws-for-multilingual-models/
- Record: https://forck.live/items/1373-atlas-practical-scaling-laws-for-multilingual-models
- Subject: Research

Google Research introduces ATLAS, new scaling laws for multilingual language models, based on 774 training runs across 10M-8B parameter models and data spanning 400+ languages, providing guidance on optimal model size, data volume, and language mixtures.

### Hugging Face — Architectural Choices in China's Open-Source AI Ecosystem: Building Beyond DeepSeek

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-27T15:01:45+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment-blog-2
- Record: https://forck.live/items/1557-architectural-choices-in-china-s-open-source-ai-ecosystem-building-beyond
- Subject: Platform

StepFun released Step-Audio-R1.1, a speech-to-speech model that boasts state-of-the-art performance, beating proprietary models.

### Vercel — AGENTS.md outperforms skills in our agent evals

- Company: Vercel (vercel.com)
- Announced: 2026-01-27T13:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/agents-md-outperforms-skills-in-our-agent-evals
- Record: https://forck.live/items/20760-agents-md-outperforms-skills-in-our-agent-evals
- Subject: v0 / AI SDK / AI Gateway

Vercel evaluated two approaches for teaching coding agents framework-specific knowledge for Next.js 16 APIs. A compressed 8KB docs index embedded directly in AGENTS.md achieved a 100% pass rate, while skills maxed out at 79% even with explicit instructions telling the agent to use them. Without those instructions, skills performed no better than having no documentation at all.

### Vercel — How Sensay went from zero to product in six weeks

- Company: Vercel (vercel.com)
- Announced: 2026-01-27T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-sensay-went-from-zero-to-product-in-six-weeks
- Record: https://forck.live/items/20759-how-sensay-went-from-zero-to-product-in-six-weeks
- Subject: v0 / AI SDK / AI Gateway

Sensay built an MVP in six weeks using Vercel previews, feature flags, and instant rollbacks, avoiding the need for a DevOps team. The startup pivoted three times from an AI replica concept to an employee off-boarding platform, launching at Web Summit. CTO Marco Bettiolo estimates the company saved millions of dollars by not building internal infrastructure.

### Harvey — Harvey Academy: Shaping the Future of Legal AI Literacy

- Company: Harvey (harvey.ai)
- Announced: 2026-01-27T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/introducing-harvey-academy
- Record: https://forck.live/items/17622-harvey-academy-shaping-the-future-of-legal-ai-literacy
- Subject: Harvey

Harvey introduced Harvey Academy, a centralized learning hub offering on-demand videos and courses to build legal AI literacy. The program includes a free certification course on AI basics, platform-integrated content, and plans for role-based modules. It is designed for legal professionals, law students, and teams to learn how to apply AI in legal workflows.

### TII — Alyah ⭐️: Toward Robust Evaluation of Emirati Dialect Capabilities in Arabic LLMs

- Company: TII (tii.ae)
- Announced: 2026-01-27T10:26:42+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/tiiuae/emirati-benchmarks
- Record: https://forck.live/items/1558-alyah-toward-robust-evaluation-of-emirati-dialect-capabilities-in-arabic-llms

Hugging Face published a community article introducing Alyah, a benchmark for evaluating Arabic LLMs on the Emirati dialect. The benchmark contains 1,173 manually collected multiple-choice samples covering greetings, figurative language, poetry, and heritage. 54 models (23 base, 31 instruction-tuned) were evaluated, including Jais, Allam, Qwen, LLaMA, Fanar, and AceGPT. Accuracy on multiple-choice questions is the primary metric.

### OpenAI — PVH reimagines the future of fashion with OpenAI

- Company: OpenAI (openai.com)
- Announced: 2026-01-27T06:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/pvh-future-of-fashion
- Record: https://forck.live/items/293-pvh-reimagines-the-future-of-fashion-with-openai
- Subject: GPT / ChatGPT / API

PVH Corp., which owns Calvin Klein and Tommy Hilfiger, is adopting ChatGPT Enterprise to integrate AI into fashion design, supply chain, and consumer engagement.

### Hugging Face — Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-27T01:53:15+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/LinkedIn/gpt-oss-agentic-rl
- Record: https://forck.live/items/1559-unlocking-agentic-rl-training-for-gpt-oss-a-practical-retrospective
- Subject: Platform

A practical retrospective on training the GPT-OSS model with agentic reinforcement learning, detailing challenges such as exploding KL divergence and a fix for a log-probability mismatch in MoE architectures using the verl framework.

### Supabase — Supabase PrivateLink is now available

- Company: Supabase (supabase.com)
- Announced: 2026-01-27
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-privatelink-available
- Record: https://forck.live/items/21306-supabase-privatelink-is-now-available
- Subject: Supabase

Supabase launched PrivateLink, a capability that allows database connections to stay within the AWS network without public internet exposure. The feature is in Beta, available for Team and Enterprise plans, and supports Postgres and PgBouncer connections within the same AWS region.

### Moonshot AI — Kimi K2.5

- Company: Moonshot AI (moonshot.ai)
- Announced: 2026-01-27
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.kimi.ai/blog/kimi-k2-5
- Record: https://forck.live/items/4626-kimi-k2-5
- Subject: Kimi

Moonshot AI introduces Kimi K2.5, an open-source multimodal model with state-of-the-art coding and vision capabilities, and a self-directed agent swarm paradigm that can orchestrate up to 100 sub-agents across 1,500 tool calls. The model is available via Kimi.com, the Kimi App, the API, and Kimi Code.

### MiniMax — A Deep Dive into the MiniMax-M2-her

- Company: MiniMax (minimax.io)
- Announced: 2026-01-27
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.minimax.io/blog/minimax-m2-her
- Record: https://forck.live/items/3402-a-deep-dive-into-the-minimax-m2-her
- Title in English (translated by forck.live from en): A Deep Dive into the MiniMax-M2-her
- Subject: MiniMax models

MiniMax describes how they built a role-play agent for production usage, optimized over three years in Talkie/Xingye.

### Mistral AI — Terminally online Mistral Vibe.

- Company: Mistral AI (mistral.ai)
- Announced: 2026-01-27
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-vibe-2-0
- Record: https://forck.live/items/3305-terminally-online-mistral-vibe
- Subject: Mistral / Le Chat

Mistral AI announces a product called Terminally online Mistral Vibe.

### OpenAI — TRUSTBANK uses AI agents to personalize Furusato Nozei gifts

- Company: OpenAI (openai.com)
- Announced: 2026-01-27
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/trustbank
- Record: https://forck.live/items/295-trustbank-uses-ai-agents-to-personalize-furusato-nozei-gifts
- Subject: GPT / ChatGPT / API

TRUSTBANK partnered with Recursive to build Choice AI using OpenAI models, enabling personalized conversational recommendations that simplify Furusato Nozei gift discovery.

### OpenAI — Introducing Prism

- Company: OpenAI (openai.com)
- Announced: 2026-01-27
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-prism
- Record: https://forck.live/items/294-introducing-prism
- Subject: GPT / ChatGPT / API

OpenAI introduces Prism, a free LaTeX-native workspace with GPT-5.2 built in, for researchers to write, collaborate, and reason.

### Vercel — Agent skills explained: An FAQ

- Company: Vercel (vercel.com)
- Announced: 2026-01-26T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/agent-skills-explained-an-faq
- Record: https://forck.live/items/20761-agent-skills-explained-an-faq
- Subject: v0 / AI SDK / AI Gateway

Vercel published an FAQ explaining agent skills, an open format for packaging instructions, scripts, and resources that AI agents can discover and use automatically. The post describes how skills differ from MCP servers, tools, rules, and system prompts, and outlines problems such as prompt drift and instruction sprawl that skills address.

### Upstage — Solar Pro 3: Better reasoning at production scale

- Company: Upstage (upstage.ai)
- Announced: 2026-01-26
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.upstage.ai/blog/en/solar-pro-3-0127
- Record: https://forck.live/items/18456-solar-pro-3-better-reasoning-at-production-scale
- Subject: Solar Pro / Solar Mini

Upstage released Solar Pro 3, an update to Solar Pro 2, with improved reasoning accuracy and instruction-following quality while maintaining the same API interface, throughput, and serving behavior. The model shows a 52% improvement in instruction-following (IFBench 55.78) and a 30% improvement in complex reasoning (Arena Hard v2 62.5) compared to Solar Pro 2. Solar Pro 3 is available via the Upstage Console API, yupp.ai, OpenRouter, and BizRouter, and is free until March 2, 2026.

### Luma — Ray3.14 is here: Native 1080p, 3x cheaper and 4x faster.

- Company: Luma (lumalabs.ai)
- Announced: 2026-01-26
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://lumalabs.ai/news/ray3_14
- Record: https://forck.live/items/7973-ray3-14-is-here-native-1080p-3x-cheaper-and-4x-faster
- Subject: Ray / Dream Machine

Luma released Ray3.14, a new version of its video generation model, featuring native 1080p generation, 4x faster and 3x cheaper than Ray3, with improved motion consistency for Modify Video. It is available now in Dream Machine.

### Together AI — DSGym: A holistic framework for evaluating and training data science agents

- Company: Together AI (together.ai)
- Announced: 2026-01-26
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/dsgym
- Record: https://forck.live/items/2386-dsgym-a-holistic-framework-for-evaluating-and-training-data-science-agents
- Subject: Inference platform

Together AI introduces DSGym, a unified framework for evaluating and training data science agents. It integrates existing benchmarks and adds novel scientific analysis tasks (DSBio, 90 bioinformatics tasks) and Kaggle competitions (DSPredict, 92 competitions). Using DSGym, the team trained a 4B model (Qwen3-4B-DSGym-SFT-2k) on 2,000 synthetic query-trajectory pairs, achieving state-of-the-art performance among open-source models on general data science benchmarks.

### OpenAI — How Indeed uses AI to help evolve the job search

- Company: OpenAI (openai.com)
- Announced: 2026-01-26
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/indeed-maggie-hulce
- Record: https://forck.live/items/296-how-indeed-uses-ai-to-help-evolve-the-job-search
- Subject: GPT / ChatGPT / API

The post is a discussion by Indeed's Chief Revenue Officer about the role of AI in job search and recruiting, without announcing a new product, model, or specific capability.