# forck.live — 25–31 August 2025 > 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 2025-W35, 25–31 August 2025, 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/2025-W35 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Harvey — 5 Questions With Annie Shin - Company: Harvey (harvey.ai) - Announced: 2025-08-29T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/5-questions-with-annie-shin - Record: https://forck.live/items/17681-5-questions-with-annie-shin - Subject: Harvey Annie Shin, Engineering Manager of the New Verticals team at Harvey, describes the company's engineering culture as customer-obsessed, with a focus on full ownership and bias for action. She explains that Harvey embeds former lawyers (Applied Legal Researchers) into product teams for immediate feedback, and that engineers proactively read customer feedback channels to address needs. The New Verticals team develops novel product offerings, including a Word Add-In that brings Harvey's AI capabilities into Microsoft Word. ### Cua — When Agents Need Human Wisdom - Introducing Human-In-The-Loop Support - Company: Cua (cua.ai) - Announced: 2025-08-29 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/human-in-the-loop - Record: https://forck.live/items/16832-when-agents-need-human-wisdom-introducing-human-in-the-loop-support - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua added human-in-the-loop support to its agent platform, letting users take over any agent configuration through a web UI, generate training demonstrations by performing tasks themselves, and validate agent performance with human expertise. The update is designed for creating training data, evaluating ground-truth scenarios, and intervening when automation needs guidance. ### Amp — Bring Your Own Tools - Company: Amp (ampcode.com) - Announced: 2025-08-29 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/toolboxes - Record: https://forck.live/items/7807-bring-your-own-tools - Subject: Amp Amp now allows users to provide tools to the system via executable files placed in a directory pointed to by the AMP_TOOLBOX environment variable. Amp will invoke these executables to describe themselves and execute tool calls when the agent decides to use them. ### Harvey — Harvey Launches Academic Program with Law Schools at Stanford, NYU, Michigan, UCLA, The University of Texas, and Notre Dame - Company: Harvey (harvey.ai) - Announced: 2025-08-28T13:30:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/harvey-launches-academic-program-with-law-schools - Record: https://forck.live/items/17682-harvey-launches-academic-program-with-law-schools-at-stanford-nyu-michigan - Subject: Harvey Harvey launched a law school alliance program with six US law schools, including Stanford, NYU, Michigan, UCLA, UT Austin, and Notre Dame. The program embeds Harvey's generative AI platform into law school curricula, providing students, faculty, and administrators access to the platform and co-creating AI learning content for the 2025-26 academic year. ### Vercel — Preparing for the worst: Our core database failover test - Company: Vercel (vercel.com) - Announced: 2025-08-28T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/preparing-for-the-worst-our-core-database-failover-test - Record: https://forck.live/items/20830-preparing-for-the-worst-our-core-database-failover-test - Subject: v0 / AI SDK / AI Gateway Vercel performed a full production failover of its core control-plane database from Azure West US to East US 2 on July 24, 2025, with zero customer impact. The test followed two major Azure West US power failures in 2024 that had prevented critical write operations. Preparation involved 57 staging failovers and revealed issues such as proprietary Cosmos DB clients lacking failover awareness and hardcoded delays that were subsequently reduced from five minutes to five milliseconds. ### OpenAI — Supporting nonprofit and community innovation - Company: OpenAI (openai.com) - Announced: 2025-08-28T05:00:00+00:00 - Category: funding-company-news - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/supporting-nonprofit-and-community-innovation - Record: https://forck.live/items/484-supporting-nonprofit-and-community-innovation - Subject: GPT / ChatGPT / API OpenAI announces a $50 million People-First AI Fund to support U.S. nonprofits in scaling their impact with AI, with applications open from September 8 to October 8, 2025, for grants in education, healthcare, research, and other areas. ### Cua — Cua × HUD - Evaluate Any Computer-Use Agent - Company: Cua (cua.ai) - Announced: 2025-08-28 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/hud-agent-evals - Record: https://forck.live/items/16833-cua-hud-evaluate-any-computer-use-agent - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua announced an integration with HUD, an evaluation platform for computer-use agents, allowing users to benchmark any GUI-capable agent on real computer-use tasks. The integration supports one-line evaluations on benchmarks like OSWorld and SheetBench for models from OpenAI, Anthropic, Hugging Face, and composed GUI models, with live traces available at app.hud.ai. ### Microsoft AI — Two in-house models in support of our mission | Microsoft AI - Company: Microsoft AI (microsoft.ai) - Announced: 2025-08-28 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://microsoft.ai/news/two-new-in-house-models/ - Record: https://forck.live/items/10973-two-in-house-models-in-support-of-our-mission-microsoft-ai - Subject: MAI Microsoft AI previewed two in-house models: MAI-Voice-1, a speech generation model now powering Copilot Daily and Podcasts and available in Copilot Labs, and MAI-1-preview, a mixture-of-experts foundation model trained on ~15,000 NVIDIA H100 GPUs, now in public testing on LMArena and available to trusted testers via API access. MAI-1-preview is designed for instruction-following and everyday queries, and will be rolled out for certain text use cases within Copilot over the coming weeks. ### Microsoft AI — Two in-house models in support of our mission - Company: Microsoft AI (microsoft.ai) - Announced: 2025-08-28 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://microsoft.ai/news/two-new-in-house-models - Record: https://forck.live/items/8438-two-in-house-models-in-support-of-our-mission - Subject: MAI Microsoft AI released MAI-Voice-1, a speech generation model, and MAI-1-preview, a mixture-of-experts foundation model. MAI-Voice-1 is available in Copilot Daily, Podcasts, and Copilot Labs. MAI-1-preview is available for public testing on LMArena and to trusted testers via API access. ### Cognition — Build Your Own AI Data Analyst - Company: Cognition (cognition.com) - Announced: 2025-08-28 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cognition.com/blog/ai-data-analyst - Record: https://forck.live/items/7694-build-your-own-ai-data-analyst - Subject: Devin / SWE Cognition trained an on-demand data scientist that answers analytics questions in minutes instead of days or weeks. ### xAI — Grok Code Fast 1 - Company: xAI (x.ai) - Announced: 2025-08-28 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://x.ai/news/grok-code-fast-1 - Record: https://forck.live/items/3486-grok-code-fast-1 - Subject: Grok xAI introduced Grok Code Fast 1, a speedy and economical reasoning model that excels at agentic coding. ### OpenAI — Collective alignment: public input on our Model Spec - Company: OpenAI (openai.com) - Announced: 2025-08-27T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/collective-alignment-aug-2025-updates - Record: https://forck.live/items/485-collective-alignment-public-input-on-our-model-spec - Subject: GPT / ChatGPT / API OpenAI conducted a survey with over 1,000 people globally to gather public input on AI behavior, comparing their views to the company's Model Spec and using the results to align AI defaults with diverse human values. ### OpenAI — OpenAI and Anthropic share findings from a joint safety evaluation - Company: OpenAI (openai.com) - Announced: 2025-08-27T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/openai-anthropic-safety-evaluation - Record: https://forck.live/items/486-openai-and-anthropic-share-findings-from-a-joint-safety-evaluation - Subject: GPT / ChatGPT / API OpenAI and Anthropic share findings from a first-of-its-kind joint safety evaluation, testing each other's models for misalignment, instruction following, hallucinations, jailbreaking, and more. ### Cua — Agent Framework 0.4 with Composite Agents: Two Brains Are Better Than One - Company: Cua (cua.ai) - Announced: 2025-08-27 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/composite-agents - Record: https://forck.live/items/16834-agent-framework-0-4-with-composite-agents-two-brains-are-better-than-one - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua released Agent framework 0.4, which normalizes coordinate systems and output formats across GUI models from providers including Anthropic, OpenAI, Hugging Face, and OpenRouter. The update introduces composite agents that pair a grounding model for visual understanding with a planning model for reasoning, enabling models not designed for computer use, such as GPT-4o, to gain GUI capabilities. ### Together AI — DeepSeek-V3.1: Hybrid Thinking Model Now Available on Together AI - Company: Together AI (together.ai) - Announced: 2025-08-27 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.together.ai/blog/deepseek-v3-1-hybrid-thinking-model-now-available-on-together-ai - Record: https://forck.live/items/2412-deepseek-v3-1-hybrid-thinking-model-now-available-on-together-ai - Subject: Inference platform Together AI announces the availability of DeepSeek-V3.1, a hybrid model that offers both non-thinking and thinking modes, with production-ready serverless APIs and dedicated endpoints. ### Replit — Introducing the Most Comprehensive Design Support for AI Apps - Company: Replit (replit.com) - Announced: 2025-08-26T15:50:38+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/introducing-comprehensive-design-support-for-ai-apps - Record: https://forck.live/items/17944-introducing-the-most-comprehensive-design-support-for-ai-apps - Subject: Replit Agent Replit introduced design support for apps built on its platform, including themes available to all users and a beta design system import for enterprise customers. Themes allow managing colors, fonts, and UI properties across an app with a single update. Enterprise customers can import a design system from Figma or upload a package of design tokens and rules to apply to new Agent-built apps. ### Harvey — Harvey’s New Drafting Tools Meet You Where You Work - Company: Harvey (harvey.ai) - Announced: 2025-08-26T14:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/harveys-new-drafting-tools-meet-you-where-you-work - Record: https://forck.live/items/17683-harvey-s-new-drafting-tools-meet-you-where-you-work - Subject: Harvey Harvey introduced two new drafting tools: an in-app Draft Editor for generating and editing documents within Harvey, and a Word Add-In that brings Harvey's Assistant into Microsoft Word for querying and editing documents. The Draft Editor merges Draft and Assist into one experience and allows manual and AI-powered edits, while the Word Add-In offers Ask and Edit modes for clause-specific and document-wide revisions. ### OpenAI — Helping people when they need it most - Company: OpenAI (openai.com) - Announced: 2025-08-26T04:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/helping-people-when-they-need-it-most - Record: https://forck.live/items/487-helping-people-when-they-need-it-most - Subject: GPT / ChatGPT / API OpenAI discusses its approach to safety for users in mental or emotional distress, acknowledges current system limitations, and describes ongoing work to improve. ### Cua — Ubuntu Docker Support in Cua with Kasm - Company: Cua (cua.ai) - Announced: 2025-08-26 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/ubuntu-docker-support - Record: https://forck.live/items/16835-ubuntu-docker-support-in-cua-with-kasm - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua has added Ubuntu Docker support, providing a full Linux desktop inside a Docker container that is viewable in a browser. The container runs on macOS, Windows, and Linux without requiring a VM or extra clients, and is pre-configured for computer-use automation with Cua's Computer SDK. ### Harvey — Harvey Expands Local Presence in Latin America to Support Growing Legal Market - Company: Harvey (harvey.ai) - Announced: 2025-08-25T14:00:00+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/harvey-expands-local-presence-in-latin-america-to-support-growing-legal-market - Record: https://forck.live/items/17684-harvey-expands-local-presence-in-latin-america-to-support-growing-legal-market - Subject: Harvey Harvey announced the launch of a dedicated team in Mexico to support legal professionals across Latin America. The company already works with over 80 legal teams in 13 countries in the region. The local team aims to provide closer support and help firms adopt AI while meeting regional regulatory standards. ### Factory AI — The Context Window Problem: Scaling Agents Beyond Token Limits - Company: Factory AI (factory.com) - Announced: 2025-08-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://factory.com/news/context-window-problem - Record: https://forck.live/items/17826-the-context-window-problem-scaling-agents-beyond-token-limits - Subject: Droid Factory describes its approach to scaling agentic workflows by building multiple layers of scaffolding that treat context as a scarce resource, including structured repository overviews, semantic search, targeted file operations, and integrations with enterprise context sources like Datadog, Slack, and Notion. The post explains why naive vector retrieval fails for code navigation and how Factory's architecture addresses these limitations. ### Cua — Computer-Use Agents SOTA Challenge: Hack the North + Global Online - Company: Cua (cua.ai) - Announced: 2025-08-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cua.ai/blog/cua-hackathon - Record: https://forck.live/items/16836-computer-use-agents-sota-challenge-hack-the-north-global-online - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua and Ollama are organizing a hackathon at Hack the North and globally, from September 12-14, 2025, with a separate global online track open until September 28. Teams build computer-use agents using the Cua Agent Framework and are scored on OSWorld-Verified. Prizes include a MacBook Air M4, cash, and a guaranteed YC partner interview for the top team. ### Amp — MCP Permissions - Company: Amp (ampcode.com) - Announced: 2025-08-25 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/mcp-permissions - Record: https://forck.live/items/7810-mcp-permissions - Subject: Amp Amp introduced a new setting, amp.mcpPermissions, that allows users to define rules to block or allow MCP servers using pattern matching syntax. The first matching rule determines the action, and if no rules match, the MCP server is allowed by default. ### Amp — Enterprise Managed Settings - Company: Amp (ampcode.com) - Announced: 2025-08-25 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/enterprise-managed-settings - Record: https://forck.live/items/7809-enterprise-managed-settings - Subject: Amp Amp now allows system administrators to configure organization-wide managed settings that override individual settings for Enterprise Amp customers, enabling security and standardization.