Daily AI News Roundup – July 31, 2026
Microsoft just posted its most consequential AI earnings ever — and buried a warning about Copilot adoption in the footnotes. The EU is forcing Google to hand its Android keys to rivals. ByteDance launched a video model that may be the best in the world. And the EU AI Act’s user-transparency rules hit Sunday. Here’s your Thursday briefing.
Want more AI news without waiting for me to finish yelling at the latest development? Check out The SkyNet Tribune, my continuously updated AI news page featuring fresh headlines from across the web every hour. It is faster than a daily roundup, less exhausting than social media, and only slightly more likely to become self-aware.
1. Microsoft Q4: Azure Tops $100 Billion, AI ARR Hits $37B — But Copilot Has an Adoption Problem
Microsoft’s fiscal Q4 2026 results, reported Tuesday, mark a genuine milestone: Azure crossed $100 billion in annual revenue for the first time, up 41% year-on-year, and the company’s AI business reached $37 billion in annualized revenue — up 123% from a year ago. Capital expenditures jumped 69% to $41 billion in the quarter as Microsoft continues building out the data center infrastructure that OpenAI, enterprise customers, and its own Copilot products depend on. Commercial remaining performance obligations (a proxy for contracted future revenue) grew to $678 billion.
The more complicated story is Copilot. Satya Nadella announced on the earnings call that Microsoft will merge Copilot chat, GitHub Copilot, its Cowork tools, and its Autopilot agent system into a single unified “super app” launching in August — with a paid Autopilot tier for background AI agents. The move comes with a candid admission embedded in analyst notes: fewer than 4.5% of Microsoft’s 450 million commercial Microsoft 365 customers currently pay for Copilot features, and of those, only 20–30% use it on a weekly basis. At the scale Microsoft is spending — $41 billion in a single quarter — that adoption rate is a quiet crisis. The unified app is a bet that simplification will unlock the remaining 95.5%.
2. The EU Just Ordered Google to Open Android to Rival AI Assistants — and Hand Over Search Data
In a landmark pair of binding decisions, the European Commission issued two orders that will reshape how AI assistants reach consumers on two billion Android phones. The Commission ordered Google to unlock 11 Android features it currently reserves for its own Gemini assistant — including voice activation, cross-app task execution, and the ability to read and act on on-screen context — and make them available to competing AI assistants on equal terms. Users will be able to summon a third-party assistant exactly the way they now summon Gemini, and have it carry out tasks across their apps. Changes are expected to reach users by July 2027.
The second order is equally significant. Google must begin sharing anonymized search data with competing search engines and AI chatbots starting January 2027 — the kind of query-and-result training data that has given Google’s AI products a structural advantage competitors can’t replicate. Both decisions implement the EU’s Digital Markets Act, which requires dominant “gatekeeper” platforms to give rivals the same access they grant themselves. The practical effect: any AI assistant — from Claude to Perplexity to a future competitor — could become a first-class citizen on Android, with access to the same hardware features and training signals that Gemini currently monopolizes. Google has not said whether it will appeal.
3. OpenAI Is Giving 100,000 Scientists Free Access to GPT-5.6 Pro
OpenAI quietly launched one of its most consequential access programs this week: ChatGPT for Academic Researchers, which will give 100,000 scientists free access to the GPT-5.6 Sol Pro model — the equivalent of OpenAI’s $200-per-month ChatGPT Pro subscription — at no cost, as part of a $250 million commitment through 2027. The program begins with 10,000 participants this summer, each of whom can invite four collaborators from the same institution, creating a peer-expansion structure that could move quickly through research universities. Access includes expanded deep research capabilities, higher usage limits, larger context windows, and hands-on support from OpenAI engineers.
The strategic logic is clear. Researchers who build workflows and methodologies around GPT-5.6 Pro will publish with it, train students on it, and create a dependency that drives institutional adoption — the same playbook that made Google Scholar indispensable to academia and AWS indispensable to startups. Notably, model weights are explicitly off-limits: researchers get API and interface access, not the ability to inspect, fine-tune, or replicate OpenAI’s models. It’s access with guardrails — and a very long-term bet on scientific prestige as a distribution channel.
4. White House Frontier AI Framework Drops Tomorrow — Meta Is Sitting It Out
The August 1 deadline set by Trump’s June 2 executive order arrives tomorrow, and with it the voluntary framework for government pre-release review of frontier AI models. OpenAI, Anthropic, Microsoft, Google, Cloudflare, and JPMorgan Chase are all participating — giving the NSA and CISA a 30-day window to assess each new frontier model before public release, in exchange for advance notice of emerging threats. But there’s one notable absence: Meta is not in the deal. The company that has been the loudest opponent of open-weight AI restrictions apparently drew the line at voluntary pre-release government access to its models.
The White House simultaneously confirmed that Gold Eagle — the AI Cybersecurity Clearinghouse ordered by the same executive order — has been fully operational since July 14, coordinating AI-assisted vulnerability scanning, patch validation, and threat information sharing across critical infrastructure including healthcare, banking, and utilities. Together, the pre-release review framework and Gold Eagle represent the most structured US government engagement with frontier AI that has ever existed — and the most politically palatable version of AI oversight the current administration was willing to support: entirely voluntary, no mandatory licensing, no prior restraint.
5. ByteDance Drops Seedance 2.5 — 30-Second Native 4K AI Video With Localized Editing
ByteDance’s cloud services unit Volcengine launched Seedance 2.5 today in China, and the specifications push well beyond what Western video AI models currently offer. The model generates native 30-second clips at 4K resolution in a single pass — no stitching together shorter segments — and accepts up to 50 reference inputs (images, audio clips, or video snippets) in a single generation. A localized editing feature lets users redraw a specific character, object, or detail in a frame without touching anything else in the scene. A copyright filter blocks generation of recognizable real faces and copyrighted characters, a notable safety measure built into the base model.
The competitive context: Seedance 2.0, the predecessor, already ranked first on the independent Artificial Analysis Video Arena — ahead of Google Veo 3.1 and Kling 3.0, and at significantly lower cost. Seedance 2.5 widens that lead considerably on runtime and resolution. The model is available in China now via Volcengine, with a wider international rollout expected after August 3. For anyone building AI video workflows — in film, advertising, social content, or enterprise — Seedance 2.5 is now the model to benchmark against. The fact that it comes from ByteDance, which is subject to ongoing US regulatory scrutiny, will complicate enterprise adoption outside China.
6. EU AI Act Article 50 Takes Effect Sunday — 48 Hours to Get Your Disclosures Right
This Sunday, August 2, the EU AI Act’s Article 50 transparency obligations become enforceable law. If you haven’t acted yet, here’s what that means in practice: any AI system that interacts with users must clearly identify itself as AI — no more chatbots that decline to say what they are. Synthetic media (AI-generated images, audio, video) must be labeled as machine-made when intended for public use. AI-generated text designed to influence public opinion on political or social matters must be disclosed as AI-generated. These requirements apply to any company whose systems reach EU residents, regardless of where the company is headquartered.
Critically: the Digital Omnibus passed by the European Parliament in June pushed the Annex III high-risk system requirements back 16 months, to December 2, 2027 — so the employment screening tools, healthcare AI, and credit-scoring systems that had compliance teams in crisis mode get significant breathing room. But Article 50 was untouched by the omnibus. Fines go up to €15 million or 3% of global annual turnover. If you’re shipping an AI-powered product that talks to EU users, make sure it introduces itself — before Sunday morning.
That’s your AI briefing for Thursday, July 31. White House framework lands tomorrow, EU AI Act transparency rules hit Sunday, and August is shaping up to be just as busy as July. Have a great weekend.