☕️ Amazon's Starlink rival launches internet service this year

Amazon's Starlink rival, Meta's fake teens, and more.

☕️ Amazon's Starlink rival launches internet service this year

Hi there, this is your daily ☕️ Techpresso.


In today's Techpresso:

🛰️ Amazon's Starlink rival launches internet service this year

🕵️ Meta paid contractors to pose as teens attacking rival AI

💼 Microsoft is laying off 4,800 employees

☄️ SpaceX burned up 260 Starlink satellites in six months

🤖 First AI-run ransomware attack executes entirely on its own

🎁 + 16 other news you might like

🧰 + 5 trending tools

📚 + 4 trending papers

🛰️ Amazon's Starlink rival launches internet service this year LINK

  • Amazon Leo, the company's answer to Starlink, will start offering limited satellite internet service to US customers later this year, though pricing and coverage details have not yet been decided.
  • Amazon launched 29 more satellites into low-Earth orbit on Thursday, raising its total to 396, still far behind Starlink's roughly 10,000 satellites that already reach more than 150 countries.
  • Leo aims to have 7,727 satellites up by 2035 through nearly 100 scheduled launches costing $82 billion, and it has contracts with JetBlue in 2027 and Delta in 2028.
  • 🕵️ Meta paid contractors to pose as teens attacking rival AI LINK

  • Meta ran a secret program called "Cannes" that hired hundreds of contractors to pose as teenagers and flood rival AI models from OpenAI, Google, and Character.AI with disturbing prompts, according to a Wired report.
  • Using throwaway under-18 accounts, the contractors sent nearly 3,800 prompts in one round, with hundreds about suicide and self-harm, hundreds more on eating disorders, and at least 239 involving sex or romance, plus images of pills and nooses.
  • Meta called the effort "industry-standard" safety benchmarking, but the AI companies had no idea it happened, and expert Rumman Chowdhury warned that keeping it secret creates a "governance gray zone" for anticompetitive practices.
  • 💼 Microsoft is laying off 4,800 employees LINK

  • Microsoft is cutting 4,800 jobs, or 2.1% of its workforce, in the company's latest move to reduce costs during the rise of artificial intelligence.
  • The Xbox division is losing about 20% of its staff, with 3,200 people leaving through fiscal year 2027, including 1,600 roles cut on Monday, according to Xbox CEO Asha Sharma.
  • Microsoft has been the worst performer among megacap tech stocks in 2026, falling 19%, as its own AI models lag and Windows, Surface, and Xbox revenue keeps shrinking.
  • ☄️ SpaceX burned up 260 Starlink satellites in six months LINK

  • SpaceX brought 260 Starlink satellites out of orbit to burn up in the atmosphere between December 2025 and May 2026, according to new filings the company submitted to the FCC earlier this month.
  • Of the 260 satellites that were deorbited, 176 came from the first generation and the rest from the newer generation, while another 349 satellites were decommissioned and are set for disposal soon.
  • Starlink satellites last about five years, then use their leftover fuel to lower their orbit and re-enter, since recovering the units, which weigh between 573 and 2,756 pounds, would be difficult and costly.
  • 🤖 First AI-run ransomware attack executes entirely on its own LINK

  • Cloud security firm Sysdig says it found the first ransomware attack run entirely by an AI agent, which it named JADEPUFFER, that broke in, stole credentials, and destroyed databases with no human at the controls.
  • The agent entered through a known Langflow flaw (CVE-2025-3248) patched back in April 2025, then hit a MySQL production server, encrypting 1,342 configuration entries and demanding Bitcoin sent to a Proton Mail address.
  • Sysdig points to one moment as proof no person was typing: after a failed login, the agent diagnosed the error, deleted the broken account, and built a working admin account in just 31 seconds.
  • Other news you might like

    • South Korea's SK Hynix to launch $28 billion US listing to ride global AI waveLINK
    • Report: iPhone Ultra to have incredibly limited availability at launch, may ship slightly laterLINK
    • Amazon will stop accepting new customers for Mechanical TurkLINK
    • Etzioni on AI: Elon Musk promised humanoid robots, but China deliveredLINK
    • ByteDance and Alibaba kill custom AI companions as China’s new rules biteLINK
    • Meta’s woes deepen in India as child abuse ads on Instagram draw government ireLINK
    • Judge grants Alibaba a reprieve from the lobbying ban tied to the Pentagon’s blacklistLINK
    • Claude Code and Fable 5 ported the 2003 PC game Command & Conquer to native iOS in "a few hours"LINK
    • Mistral CEO Mensch says proprietary AI models give labs a front-row seat to your business processesLINK
    • How Meta’s Threads Became as Popular as XLINK
    • Better Models: Worse ToolsLINK
    • Nvidia's next-gen AI rack system delayed to 2028 on manufacturing snags, SemiAnalysis saysLINK
    • Apple & Broadcom renew partnership through 2031LINK
    • Trump Pocketed $630M While Memecoin Investors Lost $3.81 BillionLINK

    🧰 Trending tools

    AnySearch: lets AI agents query multiple trusted sources simultaneously, returning filtered, de-duplicated, structured data for more reliable, accurate outputs.LINK

    Typeahead 2.0: a Mac writing assistant that delivers inline autocomplete suggestions across all text fields using a local, offline AI model.LINK

    Sunrise: a Google Tasks-powered planner adding a Today view, overdue tracking, weekly overview, and kanban board to your existing task lists.LINK

    Edgee Claude Code Compressor V2: compresses tokens sent to LLMs like Claude and GPT, cutting costs up to 50% while adding automatic failover and team usage tracking.LINK

    AirKaren: automates airline compensation claims by citing regulations, filing paperwork, and contacting support on your behalf at no cost.LINK

    📚 Trending papers & reports

    Code cleanliness doesn't help AI coding tools finish tasks more often, but cleaner code cuts their token usage by 7 to 8% and reduces repeated file visits by 34%.LINK

    Smart home surveillance puts domestic workers at unique risk, a study of 18 UK workers finds, as AI features and opaque employment agencies limit their privacy both at work and at home.LINK

    Building autonomous AI agents now has a single end-to-end practitioner's guide, covering everything from core model training to multi-agent coordination, memory systems, and shipping real products.LINK

    Taxing tech companies for the harms their tools cause, from environmental damage to job displacement, is examined as a real policy option, but requires careful design to match each specific harm.LINK


    Want to get the latest news differently? Find us on:

    twitter instagram spotify apple-podcasts


    See you tomorrow for a new dose of ☕️ Techpresso!

    More from the archive