JadidJust-in AI Developments, Incidents & Deadlines
June 2026
SMTWTFS123456789101112131415161718192021222324252627282930

AI news, Monday, June 8, 2026

9 stories

TLDR

  • Apple WWDC: Siri rebuilt on Gemini, Claude becomes a system-level iPhone AI option (Jun 8) — iOS 27's new Extensions framework lets users set Claude (or ChatGPT) as the default AI for Writing Tools, Image Playground, and Siri — a major distribution channel for any team building AI-native apps.
  • Tim Cook's final WWDC keynote as CEO (Jun 8) — Cook steps down September 1; hardware engineering SVP John Ternus becomes Apple CEO, likely in time to unveil the foldable iPhone.
  • NVIDIA Nemotron 3 Ultra (Jun 4) — 550B open MoEMixture of expertsA model made of many sub-networks, called experts, where a learned router sends each input to only a few of them.Shazeer et al. 2017 ↗ reasoning modelA model that works through hidden reasoning steps before its final answer. Those steps are billed as output.OpenAI ↗ released; benchmarks place it as America's most capable open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ model, with 5× the throughput of competing open models.
  • GitHub Copilot switched to usage-based billing June 1 — All plans now consume GitHub AI Credits per tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗; developers reporting faster-than-expected depletion. Check your org's usage budgets.
  • EU AI Act full enforcement in 55 days (Aug 2) — The Omnibus agreement (May 7) extended some HRAIS deadlines, but the August 2 effective date for most obligations stands.

Action Items

Things your engineering team should look at or do this week.

  • Audit GitHub Copilot credit spend — As of June 1, all Copilot plans bill per-token via AI Credits ($0.01/credit). Set org-level monthly credit budgets in GitHub settings before usage surprises hit; agent-heavy workflows drain credits faster than chat. (GitHub Changelog | gHacks coverage)

  • Track iOS 27 Extensions integration — Apple's new multi-AI Extensions system ships in iOS 27 beta (released today). If your team builds iOS apps that surface AI features, Anthropic will offer an Extensions SDK — first-mover integrations become the default AI for millions of iPhones. Watch the WWDC developer session and the Apple developer documentation for the Extensions API. (Apple Newsroom | 9to5Mac)

  • Review EU AI Act compliance before August 2 — Full enforcement is 55 days out. The Omnibus (May 7) narrowed HRAIS requirements and extended some deadlines, but the core obligations (transparency for AI-generated content, consumer notices, developer documentation) take effect August 2. US companies deploying AI to EU users should check the H&K briefing. (Holland & Knight briefing | EU Council Omnibus press release)

Models & Tools

  • Apple iOS 27 AI Extensions (Jun 8) Trending Agents — Apple's WWDC keynote unveiled a rebuilt Siri running on a custom 1.2-trillion-parameter Google Gemini model (via a ~$1B/year licensing deal), using a three-tier routing architecture: on-device Apple models → Apple Private Cloud → Google Cloud Blackwell B200 GPUsGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗ for heavy reasoning. iOS 27, macOS 27, and iPadOS 27 introduce an AI Extensions system that lets users designate Claude, ChatGPT, or Gemini as the default provider for Siri, Writing Tools, and Image Playground — making Anthropic an opt-in system AI for the first time on Apple devices. iOS 27 Beta 1 released today. (Apple Newsroom | TechTimes | Tom's Guide live)

  • WebMCP: browser standard for AI agents (early preview, Jun 2026) Agents — Co-authored by Google and Microsoft, WebMCP is a proposed web standard that lets any webpage expose structured tools to AI agents — two new browser APIs (Declarative for HTML forms, Imperative for JS interactions) that eliminate DOM scraping. Now in Chrome 149 origin trial; Microsoft Edge expected to follow. Early data shows 67% fewer errors and 45% better task completion vs. visual scraping. No announcements yet from Firefox or Safari. (Chrome Developers | The New Stack)

  • NVIDIA Nemotron 3 Ultra (Jun 4) Open Source Infra — Fully open 550B-parameter Mixture-of-Experts reasoning model (55B active params) built for long-running agents; scores 48 on the Artificial Analysis Intelligence Index, making it the highest-performing open-weight model from a US lab. Ships with 5× the throughput of competing open models and is available across 25+ platforms including Cursor, CrowdStrike, and Oracle Cloud Infrastructure. Small (Nano) and medium (Super) sizes also released. (NVIDIA Newsroom | Artificial Analysis)

  • Claude Mythos: Project Glasswing expands to critical infrastructure (Jun 2) Security — Anthropic expanded access to Claude Mythos Preview — its most powerful, still-unreleased-publicly model — to 150+ organizations across 15+ countries in power, water, healthcare, communications, and hardware sectors, for AI-assisted vulnerability discovery. Anthropic says it expects to bring Mythos-class capability to general customers "in the coming weeks." On SWE-benchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ Verified, Mythos Preview leads at 93.9% — 5 points above Claude Opus 4.8. (TechCrunch | Anthropic red team blog)

APIs & Pricing

  • GitHub Copilot: all plans now on usage-based AI Credits (Jun 1) Trending — Premium Request Units (PRUs) replaced by GitHub AI Credits (1 credit = $0.01), billed on input + output + cached tokens per published API rates. Monthly credit allotments: Pro+ $39, Business $19/user, Enterprise $39/user. Code completions remain unlimited. Developers are reporting credits depleting faster than the old PRU model, particularly in agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ workflows that use claude-sonnet or GPT-5.5 as the backing model. Set per-user and org-level budget caps now. (GitHub Blog | GitHub Docs)

Industry & Business

  • Tim Cook to become Apple Executive Chairman; John Ternus named CEO effective September 1 (announced Apr, executing Jun) Trending — Cook's handoff, announced in April and underway today with WWDC, closes a 14-year era. Ternus (hardware SVP, credited with the M-series Apple Silicon strategy) is expected to unveil Apple's foldable iPhone as one of his first major product decisions. The move solidifies Apple's platform bet on AI hardware differentiation under Ternus. (Apple Newsroom | CNBC)

  • Supabase raises $500M Series F at $10.5B valuation (Jun 4) Data — Led by GIC (Singapore sovereign wealth fund), with Stripe, Salesforce Ventures, Accel, and Y Combinator participating. Supabase's valuation doubled in 8 months on the back of AI-assisted "vibe coding" — Claude Code is now cited as the single largest contributor to new database creation on their platform. Total funding exceeds $1B. Signals that agentic infrastructure (auth, storage, Postgres-as-backend) is becoming its own investable category. (CNBC | PR Newswire)

Policy & Safety

  • EU AI Act: 55 days to full enforcement (August 2) (Jun 8 countdown) Policy — The AI Act's main obligations apply in full on August 2, 2026. The May 7 Omnibus agreement extended HRAIS compliance deadlines for systems placed on market before August 2 and simplified documentation requirements for companies up to 750 employees / €150M revenue — but the core Article 50 transparency requirements (labeling AI-generated content) and adverse-decision notices remain on schedule. Companies should verify which tier their AI systems fall into; the Commission's Code of Practice on AI-generated content labeling was targeted for finalization in June. (Holland & Knight | EU Council press release)

Benchmarks & Evals

  • SWE-bench Verified leaderboard (Jun 2 snapshot) — Claude Mythos Preview leads at 93.9% (preview, not publicly available), followed by Claude Opus 4.8 at 88.6% and Claude Opus 4.7 (Adaptive) at 87.6%. On SWE-bench Pro (considered less contaminated), Mythos leads at 77.8%. Context: the gap between SWE-bench Verified scores (~90%) and Pro scores (~78%) is being attributed to training-data contamination in the Verified set — teams benchmarking coding agents should run on Pro for a more honest signal. (BenchLM.ai | Scale SWE-Bench Pro)

Type to search · ↑↓ to move · enter to open