Skip to content
Daily Brief News Daily Brief

9 AI stories from June 25, 2026: Gemini 3.5 Pro delayed to July, Figma ships code layers and Motion, ByteDance Seedance 2.5, GPT-4.5 retirement Saturday, Meta Arena prediction market, Conmy joins Anthropic, Fable 5 Day 13, EU AI Act 38 days out, and GPT-5.6 still unconfirmed

· by Pondero Newsdesk · 9 stories

AI news daily brief: 2026-06-25

Nine stories today: a Google model delay that breaks a public CEO promise, a major Figma platform overhaul, a ByteDance video breakthrough, two OpenAI product timeline items, a Meta competitor launch ordered directly by Zuckerberg, a third Google AI defection to a frontier rival, and two regulatory countdowns operators need on their calendars.

Gemini 3.5 Pro slips to July as Google misses Sundar Pichai's I/O commitment

Google pushed Gemini 3.5 Pro's general availability to July 2026, per Business Insider citing a person familiar with the matter, corroborated by Crypto Briefing and Startup Fortune. The delay breaks an on-stage commitment Sundar Pichai made at Google I/O on May 19 ("give us until next month"). The model remains in limited Vertex AI enterprise preview with a 2-million-token context window and a Deep Think reasoning mode gated to Ultra subscribers. The slip comes as Alphabet stock absorbed the back-to-back exits of AlphaFold lead John Jumper and Transformer co-author Noam Shazeer to rival labs, per the Gemini API changelog. Teams evaluating Gemini 3.5 Pro for production workloads should treat July as the earliest realistic window and hold off on pricing commitments until GA is confirmed.

Full story: Gemini 3.5 Pro delayed to July as Google misses Pichai's I/O promise

Figma Config 2026: code layers, Motion timeline, and generative AI plugins arrive

Figma announced a platform overhaul at Config 2026 on June 24, per the official blog. Code layers let designers convert any design layer into an interactive code layer with one click or a prompt. Figma Motion adds a keyframe timeline with export in CSS, JSON, React, MP4, WebM, animated SVG, and GIF. An AI-generative plugin builder takes text prompts and turns them into repeatable skills for Figma's AI agent. Weave tools bring AI image generation, shader fills, and video editing into the canvas. Design teams that route production assets through Figma now have a path to skip separate animation tools; the CSS/React export targets the direct handoff gap that has pushed some teams toward ProtoPie or Framer.

Full story: Figma Config 2026 ships code layers, native Motion, and AI-generative plugins

ByteDance Seedance 2.5: native 30-second 4K video in one generation pass, launching early July

ByteDance announced Seedance 2.5 at the Volcano Engine 2026 conference on June 23, per The Next Web, with public launch targeted for early July. The defining change is native 30-second 4K video generation in a single pass. Previous models required stitching multiple 6-second or 10-second clips, which produced seam artifacts at join points. The model also accepts up to 50 all-modality reference inputs simultaneously, up from 12 in Seedance 2.0. Enterprise beta is live now. For video production teams currently using Sora or Runway for long-form clips, this is the first native alternative worth evaluating on raw generation length before the July launch date.

Full story: ByteDance Seedance 2.5 targets Sora with native 30-second 4K generation

GPT-4.5 exits ChatGPT on Saturday as OpenAI's consumer lineup goes fully GPT-5.x

GPT-4.5 will be retired from ChatGPT on June 27, 2026, closing the 30-day sunset period announced May 28, per OpenAI's Model Release Notes. The retirement applies to ChatGPT only; the API retains access. Developers routing GPT-4.5 calls through ChatGPT-facing workflows lose access Saturday. With that removal, ChatGPT's model lineup becomes entirely GPT-5.x. OpenAI o3 follows on August 26 (90-day sunset). Saturday is the live deadline for anyone with ChatGPT-dependent automations that explicitly target GPT-4.5.

Full story: OpenAI retires GPT-4.5 from ChatGPT Saturday

Zuckerberg orders Meta to build Arena, an AI-powered play-money prediction market

Mark Zuckerberg instructed a team to build Arena, a standalone app where users receive a daily virtual allotment of play money to bet on future-event outcomes, per NPR and TechCrunch. Llama generates questions from trending topics and resolves markets in near-real-time. The play-money model avoids direct CFTC regulation. DraftKings and Robinhood stock fell on the news. For operators tracking Meta's AI product surface, Arena is the clearest signal yet that Meta is building consumer AI applications around Llama rather than leaving that ground to OpenAI and Anthropic.

Full story: Zuckerberg orders Meta to build Arena AI prediction market app

Arthur Conmy joins Anthropic from Google DeepMind, the third major departure in seven days

Arthur Conmy announced on June 25 that he is joining Anthropic to work on train-time alignment for upcoming models, per his personal site. He spent three years at Google DeepMind working on post-training to improve Gemini's alignment behavior. His page now reads "From 2023-2026, I worked at Google DeepMind." Conmy follows Noam Shazeer (to OpenAI) and John Jumper (to Anthropic) in a seven-day run of notable Google AI researcher exits. Three alignment-credentialed researchers leaving for competitors in one week is a talent signal worth tracking: each arrival strengthens safety and research capacity at the destination lab while thinning it at Google.

Full story: Arthur Conmy joins Anthropic from DeepMind as Google loses third AI researcher in seven days

Fable 5 ban enters Day 13 with July 8 biometric verification deadline as the first restoration trigger

As of June 25, day 13 of the US government's export-control directive, Claude Fable 5 and Claude Mythos 5 remain offline globally, per Anthropic's statement page. API calls to claude-fable-5 still return errors. Prediction markets hold 57% odds of restoration before July 1. Two structural milestones ahead: July 8, when Anthropic's updated privacy policy takes effect and introduces biometric ID verification via Persona, and August 1, the 60-day executive order deadline for the covered frontier model framework. Arthur Conmy's hire today signals Anthropic is actively building during the ban. Operators should hold alternate model fallbacks (claude-opus-4, claude-sonnet-4) in production until a restoration date is confirmed.

EU AI Act Article 50 transparency rules take effect in 38 days, 78% of organizations still unprepared

From August 2, 2026, organizations deploying AI systems in the EU must comply with Article 50 transparency obligations, per the EU AI Act implementation timeline. The rules require users to be informed when interacting with a chatbot, AI-generated content to be marked as such, and deepfakes to be disclosed. Fines reach 35 million euros or 7% of annual global turnover. Sidley published detailed compliance guidance on June 24 noting that 78% of organizations have not taken meaningful compliance steps as of April 2026. At 38 days out, the minimum viable step is auditing every user-facing AI touchpoint for disclosure labeling before the deadline.

GPT-5.6 remains unconfirmed as the Polymarket June 22-28 window closes Saturday

As of June 25, OpenAI's ChatGPT Release Notes page shows no GPT-5.6 entry, per the official notes. Polymarket traders assign 90% probability to a June 22-28 launch. Select Pro subscribers report GPT-5.6-quality responses under the GPT-5.5 Pro label, and chief scientist Jakub Pachocki described the model internally as "a meaningful improvement over GPT-5.5," per Polymarket sourcing. The prediction window closes Saturday June 28, the same day GPT-4.5 retires from ChatGPT. If a release comes before Saturday, expect it alongside the GPT-4.5 retirement announcement rather than as a standalone drop.

Sources