AI News

Top Story: The EU Orders Google to Open Android and Share Search Data With AI Rivals

 

The Tech‑Reader AI Digest

Monday, July 20, 2026

#AI #TechNews #Digest


Podcast 🎧 • Video 📽


The European Commission ordered Google to open Android to rival AI assistants and share portions of its search data. Oracle cut 30,000 jobs to fund its Stargate infrastructure commitments. Microsoft unveiled Project Perception, a multi‑model AI security platform entering a rapidly expanding category. WAICO closed its founding conference in Shanghai, and the open‑weight countdown begins: DeepSeek V4 stable July 24, Kimi K3 free weights July 27.


Story 1: The EU Orders Google to Open Android and Share Search Data With AI Rivals

What happened:
The European Commission adopted binding requirements under the Digital Markets Act ordering Google to open Android to third‑party AI assistants and to share anonymized search data with competing developers. Eligible assistants gain voice activation and cross‑app capabilities across 11 Android feature groups, subject to certification and user consent. Google must also make anonymized ranking, query, click, and view data available on fair, reasonable, and nondiscriminatory terms. Search data sharing begins January 2027; Android interoperability is due July 2027.

The remedies target Google’s two structural advantages: default placement on billions of Android devices and two decades of search behavior data. Opening Android distribution removes a barrier that has historically limited competitors regardless of model quality. Search signal access gives developers training and ranking data that previously existed only inside Google. Kent Walker, Google’s president of global affairs, argued the decisions risk undermining privacy and security. The timing coincides with Alphabet shares falling roughly 4 percent on Gemini 3.5 Pro delay reports.

Why it matters:
This is one of the most consequential regulatory actions in AI this year. It doesn’t fine Google — it alters the competitive landscape. Every assistant developer that meets certification gains a legal path onto the dominant global mobile platform. Search data access gives competitors signal they could not otherwise obtain. The 2027 deadlines give Google time to respond, but they also place a timer on advantages that were expected to offset model delays.

Aaron’s Take:
The EU put a countdown on Google’s distribution moat. Voice activation and cross‑app access on Android, plus search data available on fair terms, create a competitive opening that didn’t exist before. The deadlines give Google time to ship a strong model, and the company has the resources to do it. But the remedies land during a difficult month: a flagship delay, talent departures, and a DMA order. Watch which companies apply for Android certification first — that list will show who intends to compete for mobile AI at scale.


Story 2: Oracle Cuts 30,000 Jobs to Fund the $500 Billion Stargate Buildout

What happened:
Oracle is cutting up to 30,000 employees — about 18 percent of its workforce — to free an estimated $8–10 billion in annual cash flow for AI infrastructure. The cuts fund Oracle’s role in Stargate, the $500 billion initiative with OpenAI and SoftBank, anchored by a $300 billion five‑year cloud contract covering 4.5 gigawatts of Oracle‑built data center capacity expected to generate roughly $30 billion a year.

Reductions hit Oracle Health, cloud infrastructure, and consulting, while sparing teams building Stargate data centers. Oracle is reallocating itself toward AI infrastructure and financing the shift through restructuring. The bet is concentrated: Oracle’s future returns depend heavily on OpenAI’s growth, revenue durability, and legal posture heading into its September IPO.

Why it matters:
Oracle’s restructuring is a clear illustration of what gigawatt‑scale AI infrastructure actually costs. The capital isn’t appearing from new revenue — it’s being extracted from existing headcount and business lines. TSMC’s record $40 billion quarter confirmed demand; Oracle’s cuts show the supply‑side trade‑offs required to meet it.

Aaron’s Take:
Thirty thousand people paid for Oracle’s bet on Stargate. That belongs in the first sentence of every analysis. The strategy is coherent — a $300 billion contract with a major AI customer is a strong anchor for long‑term infrastructure revenue. The risk is equally clear: Oracle is heavily leveraged to one customer facing multiple active legal matters ahead of an IPO. If OpenAI’s revenue materializes as projected, Oracle’s move is a masterstroke. If not, Oracle has fewer employees and a large infrastructure bill. There isn’t much middle ground.


Story 3: Microsoft Project Perception Enters the AI Security Race

What happened:
Microsoft is preparing Project Perception, an AI cybersecurity platform that routes tasks across models from Microsoft, OpenAI, and Anthropic. The system scans code, cloud infrastructure, and endpoints, identifies weaknesses, explains their impact, and proposes fixes. The orchestration layer assigns each task to the most efficient model — lower‑cost models for triage and inventory, higher‑capacity models only when complex reasoning is required. Microsoft has not confirmed public availability or pricing.

The platform enters a rapidly expanding category. Microsoft’s July Patch Tuesday used AI assistance to help fix 570 vulnerabilities. Anthropic’s Project Glasswing expanded to 150 organizations across 15 countries. AI‑security acquisitions tripled year‑over‑year. JADEPUFFER demonstrated autonomous ransomware in June. Multiple vendors — Microsoft, Anthropic, OpenAI, and specialized startups — are now competing to define machine‑speed defensive tooling.

Why it matters:
AI‑powered vulnerability detection is becoming a multi‑vendor race. Microsoft’s advantage is enterprise distribution and integration. Anthropic emphasizes security‑focused reasoning. OpenAI emphasizes rapid iteration and broad developer adoption. No single vendor has a decisive lead; the category is early, and organizations benefit from multiple well‑resourced labs pushing cost, coverage, and reliability forward.

Aaron’s Take:
The notable part isn’t vendor A versus vendor B — it’s that major labs are converging on the same architecture: inexpensive models for triage, stronger models for complex reasoning, and human review for production changes. That’s what makes continuous scanning economically viable. The differentiator will be economics, integration, and reliability over time.


Story 4: WAICO Closes WAIC — And the Open‑Weight Countdown Begins

What happened:
The 2026 World Artificial Intelligence Conference closed in Shanghai after four days that included Xi Jinping’s first WAIC keynote and the signing of the World Artificial Intelligence Cooperation Organization by 29 founding nations. The conference ran more than 140 forums with over 1,100 exhibitors. Huawei demonstrated its Atlas 950 SuperPoD, claiming 6.7× Nvidia NVL144 compute. Demis Hassabis separately called for an international AI watchdog and a US‑led coalition — a statement that lands differently now that China’s version of that institution exists.

The open‑weight countdown: DeepSeek V4 stable release on July 24 ends the preview‑build churn that kept cautious enterprises from moving production workloads. Kimi K3 open weights go free on July 27, putting a high‑performing coding model into anyone’s hands with no per‑token cost. Both releases arrive 12 days before the August 1 NSA/CISA governance framework deadline.

Why it matters:
WAICO’s durability depends on whether it publishes a charter, names leadership, sets a meeting calendar, and expands membership beyond the initial 29. Xi paired the launch with an open‑source AI endorsement and pledges of assistance to developing countries — a direct appeal to nations not present at Bletchley. The Atlas 950 demonstration is notable: a domestic Chinese system claiming to outperform Nvidia’s flagship cluster is a data point the US export control architecture was designed to prevent. The claim requires verification; the demonstration occurred.

On open weights: if DeepSeek V4 stable and Kimi K3 free weights ship on schedule, late July becomes a major procurement decision point. Capable open‑weight alternatives reduce closed‑model pricing power, and routing habits formed during such windows tend to persist.

Aaron’s Take:
Hassabis calling for a US‑led AI watchdog on the same day WAICO closed with 29 members captures the contrast between intention and institution. WAICO has a headquarters, a signing ceremony, and member states. The US has a voluntary standards framework due August 1. Institutions are built by showing up with structure. The open‑weight releases are the more immediate story — run your workloads against DeepSeek V4 stable and K3 Max when the weights drop, compare costs, and let the numbers decide.


Quick Hits — The Rest of Today’s AI World

Anthropic / Claude

  • Project Perception uses a mix of Microsoft, OpenAI, and Anthropic models for different task types. Anthropic’s models handle certain high‑complexity reasoning tasks alongside Microsoft’s and OpenAI’s strongest offerings.
  • Claude Fable 5 free access window closed July 19; credits‑based access now applies. The shift coincides with Kimi K3’s upcoming free open‑weight release.
  • Project Glasswing at 150 organizations across 15 countries. Claude for Government beta active. IPO investor meetings underway for ticker ANTH.
  • Sonnet 5 introductory pricing runs through August 31, competing directly with offerings from OpenAI, Google, and Microsoft.

OpenAI

  • Oracle’s restructuring funds a $300 billion contract with OpenAI.
  • Apple trade secret lawsuit deepened last week, naming specific staff.
  • NYT sanctions motion, Apple lawsuit, and government‑stake disclosure obligations active ahead of the September IPO.
  • GPT‑5.6 Sol, Terra, and Luna remain in production with no major incident reports.

xAI / SpaceX

  • Grok 4.5 EU regulatory clearance pending.
  • SpaceX $60 billion Cursor acquisition integration continues.

Google Gemini

  • EU DMA order — Android opening and search data sharing.
  • Gemini 3.5 Pro: fourth missed deadline; July 24 is next internal target.
  • Gemini Notebook now includes a secure cloud computer for native code execution.
  • Gemini AI Mode in Search expanded with Instacart, Canva, and YouTube Music integrations.
  • Hassabis called for an international AI watchdog.

Microsoft

  • Project Perception — full story above.
  • Microsoft Frontier Company continues enterprise deployment ramp.

Oracle / Stargate

  • 30,000 jobs cut to fund Stargate — full story above.

Meta

  • Parent distress alert feature now live.
  • Custom chip manufacturing begins September; 14 GW compute targeted by 2027.

SAP / Prior Labs

  • SAP completed its acquisition of Prior Labs and committed over €1 billion to scale it into a European frontier lab focused on structured‑data AI.

Nvidia / Hardware

  • Huawei Atlas 950 SuperPoD demonstrated at WAIC claiming 6.7× NVL144 performance.
  • SK Group Chair Chey Tae‑won warned of severe HBM shortages into 2027.
  • Etched reportedly in discussions valuing the company up to $20 billion.

Hugging Face

  • During last week’s agentic intrusion, Hugging Face’s blue team used GLM‑5.2 after commercial API guardrails blocked analysis requests — highlighting operational gaps created by safety filters.

AI Safety / Healthcare

  • RadLE 2.0 benchmark found radiology models often deliver confident but incorrect diagnoses.
  • AI text detectors fail on style imitation; scientific writing most vulnerable.

China AI — Open‑Weight Countdown

  • Kimi K3 open weights: July 27 — 2.8T parameters, free to self‑host.
  • DeepSeek V4 stable: July 24 — mandatory API migration deadline same day.
  • WAICO — 29‑nation governance body; charter and leadership will determine trajectory.
  • Huawei Atlas 950 — demonstrated at WAIC; claim unverified.
  • Apple Intelligence in China — approved with Alibaba Qwen as primary partner.

Cohere / Aleph Alpha

  • Proposed $20 billion merger remains in regulatory review.

That’s your AI world for Monday. See you tomorrow — Aaron





Aaron Rose is a software engineer and technology writer at tech-reader.blog

Catch up on the latest explainer videos, podcasts, and industry discussions below.


Popular posts from this blog

Insight: The Great Minimal OS Showdown—DietPi vs Raspberry Pi OS Lite

Running AI Models on Raspberry Pi 5 (8GB RAM): What Works and What Doesn't