AI Daily Brief — 19 June 2025
Juneteenth in the US, so the day’s flow split between holiday-quiet US lab feeds and a few sharp on-the-date stories. OpenAI confirmed to Bloomberg and Fortune that it is unwinding its Scale AI relationship after Meta’s $14.3B investment and Alexandr Wang’s defection — Google was reportedly doing the same. Anthropic added remote MCP server support to Claude Code, removing local-setup friction for any vendor URL. Midjourney’s V1 video model opened up to wider subscribers, and the BBC flagged an AI-propaganda “tsunami” around the Israel-Iran war.
Top stories
- OpenAI phases out Scale AI partnership. OpenAI confirmed to Fortune and Bloomberg it is winding down its long-running data-labeling relationship with Scale, the vendor that helped train models from GPT-4 onward. Google was reportedly cutting ties at the same time. The move follows Meta’s $14.3B deal for a 49% Scale stake and Alexandr Wang’s move to Meta. via Fortune
- Claude Code gets remote MCP server support. Anthropic announced Claude Code now connects to any remote Model Context Protocol server with no local setup — paste a vendor URL, native OAuth/SSE/stdio transports supported. Sentry and Linear shipped as reference integrations. via Anthropic
- Gemini 2.5 GA rolls through Vertex endpoints. Per Vertex AI release notes, gemini-2.5-flash-preview-04-17 began serving the May 20 GA Flash build and the preview Pro endpoints flipped to the June 5 GA Pro build on June 19 — the operational follow-through to the June 17 announcement. via Google Cloud Blog
- Midjourney V1 opens to wider subscribers. The image-to-video model, announced June 18, became broadly accessible from the $10/month Basic plan up. Pro and Mega plans get unlimited “Relax” mode video. Each video job runs roughly 8x an image job. via TechCrunch
- BBC Verify flags AI-propaganda “tsunami” around Israel-Iran war. Day 7 of the conflict drew a torrent of AI-generated fake videos, images and casualty claims. Carnegie later described the episode as the first conflict where weaponized generative-AI propaganda was central to the information battle. via Carnegie Endowment
Who shipped
Anthropic shipped remote MCP support for Claude Code. OpenAI shipped a confirmation, not a product — winding down Scale AI work. Google rolled GA Gemini 2.5 endpoints through Vertex. Midjourney widened V1 video access.
Open-source pulse
Juneteenth and the post-MiniMax-M1 lull meant no major DeepSeek, Qwen, Kimi or Hunyuan release on the day. The week’s open-weight conversation kept centering on MiniMax-M1, Mistral Magistral and Alibaba’s April Qwen3 family.
Money, infra & hardware
Helsing’s €600M Series D at ~€12B valuation from two days earlier — led by Daniel Ek’s Prima Materia with Lightspeed, Accel, Plural, General Catalyst, SAAB and new entrants BDT & MSD Partners — remained the dominant European defense-AI funding story, reframing “sovereign AI” financing mid-conflict. AMD’s MI350 unveil kept driving silicon coverage. OpenAI’s June 10 o3 price cut (80% to $2/$8 per million tokens) and o3-pro launch at $20/$80 also continued to shape the API pricing conversation.
Quiet corners
No US frontier-lab product launches on Juneteenth itself — no OpenAI, Anthropic, xAI, Meta AI or Apple Intelligence on-the-date posts beyond the Scale AI confirmation and Anthropic’s MCP announcement.
By the numbers
- $14.3B — Meta’s deal for a 49% Scale AI stake (triggers OpenAI’s wind-down)
- OAuth / SSE / stdio — transports supported by remote MCP in Claude Code
- Sentry, Linear — reference remote-MCP integrations shipped with the announcement
- $10/month — Midjourney entry plan with V1 video access
- Day 7 — Israel-Iran conflict, with weaponized AI propaganda central to the information battle
Compiled by AI Feed’s editor from verified web sources for 19 June 2025.