AI Daily Dev — June 2026

Microsoft Cancels Internal Claude Code Licenses After Burning a Year of AI Budget in Months

Affects engineers across Experiences & Devices — the unit that builds Windows, Microsoft 365, Teams, Outlook and Surface; migration to GitHub Copilot CLI by June 30.
Cause is unit economics, not quality: token-based billing consumed the full annual AI budget in months because engineers used Claude Code constantly.
Hits HN at #2 with 418 points and 398 comments — the comments treat it as proof that enterprise coding-AI pricing doesn't work at current token rates, not that Copilot won on merit.
Echoed by Uber CTO disclosing the same pattern — full 2026 AI budget gone in four months.

industry windowscentral.com

Headline Build 2026 keynote announcement: Microsoft's first homegrown frontier-tier coding model, replacing GPT-4 Turbo as the GitHub Copilot default in August 2026.
Mixture-of-experts with language-specific sub-modules; Microsoft claims it beats GPT-4 Turbo on HumanEval and MBPP, with the biggest gains in Rust and Haskell.
Adds chain-of-thought and tree-of-thought search at inference; preview demo migrated a .NET Framework app to .NET 9 end-to-end, including CI/CD.
Microsoft is explicit about the target: Claude Code overtook Copilot in developer adoption and Polaris is built to take that ground back — automatic migration plus a 3-month GPT-4 fallback for teams that opt out.

models microsoft.com

June 1 filing — beats OpenAI to the IPO queue four days after Anthropic closed a $65B Series H at a $965B post-money valuation, surpassing OpenAI's $852B.
Run-rate revenue crossed $47B earlier in May; WSJ reports Anthropic projects break-even by 2028, two years ahead of OpenAI's 2030 target.
Confidential submission lets Anthropic work through SEC review privately; share count and price are not yet set.
HN thread on the filing front-pages; top comments focus on NASDAQ's new fast-change rule (15-day eligibility for index inclusion) and S&P-index pushback over AI valuations.

industry anthropic.com

Signed June 2 — AI companies are asked (not required) to share covered frontier models with the federal government for up to 30 days before release.
Earlier May draft set a 90-day window; lobbying by Musk and Zuckerberg shortened it, and the final text explicitly bars any mandatory licensing or preclearance.
Directs NIST to author the voluntary framework and stands up an 'AI cybersecurity clearinghouse' to share vulnerability info across agencies.
HN front page (Trump signs downsized AI order after weeks of reversals): voluntary scheme + 'trusted partner' carveouts dominate the comment thread.

industry cnbc.com

Headline of OpenAI's June 2 'Intelligence at Work' livestream: Codex is being repositioned from a coding agent to a generalist white-collar workhorse.
Six role plugins (sales, data analytics, creative, product design, public-equity investing, investment banking) bundle 62 business apps and 110 prebuilt skills.
'Sites' spins up OpenAI-hosted dashboards and internal web apps from a prompt; 'Annotations' lets users highlight a region of a doc, sheet, or slide and edit it in place.
OpenAI now claims 5M weekly Codex users, with non-developers ≈20% of the base and growing 3× faster than engineers; Codex rolls into the ChatGPT app 'in the next few weeks'.

tools openai.com

Anthropic studied 832 accounts banned for malicious cyber activity between March 2025 and March 2026, mapping 13,873 actions across 482 unique ATT&CK techniques.
Medium- or high-risk actors using AI for cyber ops jumped from 33% to 56% across the year — a 1.7× rise the report frames as a structural shift, not a blip.
Malware development dominates: 67.3% of the 832 actors used Claude to build malicious software; only 6.5% reached more advanced 'lateral movement' help.
Subset of findings ships in Verizon's 2026 DBIR; Anthropic also publishes an interactive LLM ATT&CK Navigator at red.anthropic.com.

research anthropic.com

June 2 Build keynote: MAI-Code-1-Flash is a 5B-parameter coding model rolling out to all GitHub Copilot tiers (Free → Max) starting in VS Code.
Microsoft reports 51.2% on SWE-Bench Pro vs Claude Haiku 4.5's 35.2%, with up to 60% fewer tokens on harder Verified problems via 'adaptive solution length'.
MAI-Thinking-1 is a 35B-active-parameter sparse MoE reasoner with 256K context, trained entirely on commercially licensed data (no third-party distillation), in private preview on Foundry.
HN thread headlines the production-harness training story — model was tuned inside the live Copilot loop, not benchmarked into it after the fact.

models microsoft.ai

June 1 launch: first open-weight model to combine frontier coding, 1M-token context and native multimodal (text/image/video) input in one architecture.
Scores 59.0% on SWE-Bench Pro — ahead of GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%), still behind Claude Opus 4.8 at 69.2%.
New MiniMax Sparse Attention (MSA) cuts per-token compute to ~1/20 of M2 at 1M context, with 9.7× faster prefill and 15.6× faster decode.
API live now at $0.60/$2.40 per 1M input/output tokens — roughly 8–20% of leading proprietary U.S. models; weights and technical report promised within 10 days.

models minimax.io

WWDC keynote unveils a new Extensions API: users pick any AI app from the App Store as the engine for Siri, Writing Tools and Image Playground.
Default Siri now runs on a custom 1.2T-parameter Gemini model — Apple confirmed the ~$1B/year Google deal on stage.
Foundation Models framework expands so third-party iOS 27 apps can also call Apple's on-device LLM and route to chosen extensions.
Tim Cook's last WWDC keynote as CEO; iOS 27 / iPadOS 27 / macOS 27 dev betas shipped same afternoon, public release in September.

industry tomsguide.com

WWDC Platforms State of the Union (June 8) — Foundation Models framework gains image input, cloud-model routing and Dynamic Profiles; will be open-sourced later this year to run wherever Swift runs, including Linux servers.
App Intents 2.0 + Siri Extensions become the mandatory integration surface for Siri AI; SiriKit (since 2016) gets a formal deprecation notice.
Xcode 27 ships an Agent Client Protocol (ACP) client and Xcode Agents that can run tests, Playgrounds and Simulator apps; Device Hub replaces the standalone Simulator.
New Core AI framework lands as the OS-level path for on-device models on Apple silicon; an `fm` CLI and Python SDK expose the same on-device model outside Swift.

frameworks developer.apple.com