Fired OpenAI researchers deny misconduct

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

⚠️ Fired OpenAI researchers deny misconduct

📡 SpaceX announces plan to become a ‘major mobile carrier’

😇 Anthropic bans cruelty toward Claude

🛑 US bars Microsoft from green card program

🚫 Meta bans TikTok ads across its apps

📱 Apple cuts iPhone 18 Pro production due to ‘soft demand’

Plus: 🎁 11 more stories you might like, 🧰 6 tools, and 📚 5 papers.

⚠️ Fired OpenAI researchers deny misconduct LINK

• Three OpenAI safety researchers OpenAI fired last week, Jasmine Wang, Tomek Korbak, and Mikita Balesni, published an open letter denying they mishandled sensitive data and warning the dismissals are scaring colleagues away from safety work.
• OpenAI says the three were fired after an investigation found a "pattern of misconduct" that went beyond sharing information with an outside AI evaluation group, though it declined to name which specific policies they allegedly broke.
• Wang said on X she was fired for opening an executive's email by mistake, access she claims OpenAI gave her for recruiting and that IT failed to remove when she asked, calling the stated reasons suspicious.

📡 SpaceX announces plan to become a ‘major mobile carrier’ LINK

• SpaceX wants to turn its Starlink Mobile service into a major US carrier, buying low-band spectrum licenses to build a network that rivals T-Mobile, AT&T, and Verizon once the FCC approves the deal.
• The purchase gives SpaceX up to 14 megahertz of paired spectrum in the 800 MHz band, which the company says will let Starlink Mobile signals reach through walls and into buildings, adding to 2GHz spectrum it earlier bought from EchoStar.
• The FCC also cleared SpaceX to launch 15,000 V2 Starlink Mobile satellites offering more than 100 times the bandwidth of the current fleet, letting the firm mix satellite and ground-based spectrum to cover cellular dead zones.

😇 Anthropic bans cruelty toward Claude LINK

• Anthropic has rewritten its usage rules to block people from being needlessly cruel or abusive toward its Claude AI system, wading into a wider argument over whether artificial intelligence could ever be conscious.
• The new policy bans "sustained and needless abusive or cruel behavior" toward the models, and says Claude's own power to end a conversation will stay the main way the rule gets enforced.
• The move builds on a feature from last year that let Claude walk away from "persistently harmful or abusive" chats, and follows CEO Dario Amodei saying in February he was unsure whether AI models could be conscious.

🛑 US bars Microsoft from green card program LINK

• The Trump administration has suspended Microsoft, Adobe, and several other tech firms from a program that helps skilled foreign workers gain permanent residency in the US, accusing the companies of fraud.
• Labor Secretary Keith Sonderling said the government will no longer accept new or pending permanent labor certification applications tied to the suspended firms, which also include Capgemini, Cognizant, HCL, Infosys, Tata, and Wipro.
• Vice President JD Vance told Microsoft to "hire great American workers," pointing at H-1B visas for skilled jobs, nearly three-quarters of which go to workers from India, according to The Associated Press.

🚫 Meta bans TikTok ads across its apps LINK

• Meta has blocked all advertisements from ByteDance, TikTok's Chinese parent, across Facebook and Instagram in the United States and a few other countries, calling it normal practice to deny promotion to a direct rival.
• The ban started immediately in the US, Canada, Egypt, Indonesia, Japan, Thailand and Vietnam, and also covers outside advertisers running campaigns that link to TikTok or other ByteDance properties in those places.
• Meta said it won't run ads from a competitor trying to pull people off its apps, while TikTok still runs in the US as a majority American-owned joint venture serving more than 200 million users there.

📱 Apple cuts iPhone 18 Pro production due to ‘soft demand’ LINK

• Apple has told suppliers to scale back production of iPhone 18 Pro parts because demand has fallen short of its expectations, according to a report from Nikkei Asia.
• Component orders for October were reportedly cut by at least 15 percent, a move the report links to rising memory chip costs that pushed the iPhone 18 Pro and Pro Max to $1,199 and $1,299.
• Both phones keep 12GB of RAM, the same as last year's iPhone 17 Pro models, after Apple reportedly dropped a plan for 16GB to avoid the added expense of pricier memory.

Other news you might like

• R.I.P. Margaret Hamilton, whose code saved the Apollo 11 Moon landingLINK

• IDC: Worldwide PC shipments in Q3 fell 20 percent from last yearLINK

• EXCLUSIVE: AI surveillance startup Flock Safety to cut several hundred jobs amid backlash, sources sayLINK

• OpenAI caught Russians and Iranians using ChatGPT for influence campaignsLINK

• Amazon builds 1,000th satellite, is weeks away from space Internet rolloutLINK

• Introducing the Anthropic Cyber MissionLINK

• OpenAI’s revenue is reportedly $20 billion less than previously projectedLINK

• The State of AI Report 2026LINK

• Tim Cook opens up about life after stepping down as Apple CEOLINK

• SpaceX calls for better coordination in orbit after near-misses with StarlinkLINK

• Apple Set to Debut Touch-Screen MacBook and New iPad Mini in Late OctoberLINK

🧰 Trending tools

Staffcoder: practice real engineering work with 1000+ challenges across 39 paths covering React, Django, Docker, and Kubernetes in a full browser IDE, no setup needed.LINK

AgentSDR: open-source AI sales workspace for email, LinkedIn, and WhatsApp outreach with data enrichment, automations, and a self-hosted CRM that classifies replies.LINK

Comcent: browser-based voice platform connecting Twilio or SIP providers with call flows, queues, AI transcription, sentiment scoring, and voice bots for calling teams.LINK

Microsoft Surface Laptop Ultra: a 15-inch laptop with RTX Blackwell graphics and up to 128GB memory, running AI models up to 120B parameters locally without the cloud.LINK

Cakie: a local-first desktop knowledge base combining Markdown and Excalidraw canvases, with plain-text storage and end-to-end encryption for private, smooth note-taking.LINK

Melete: deploys AI agents that research, draft, and act across Slack, Notion, Linear, and GitHub, using scoped memory and requiring human approval.LINK

📚 Trending papers & reports

App-building AI agents get a new test turning design mockups into working apps, revealing the best systems deliver functional results only ~55% of the time on web and ~38% on mobile, with controls often unresponsive.LINK

Algorithmic reinvention tests show that when top AI models are blocked from recalling classic computer science algorithms, they fail to rederive nearly half of them, revealing weak capacity for genuine foundational innovation versus memorization.LINK

Code copying detector traces whether an AI coding assistant lifted snippets verbatim from its training data, matching older plagiarism checkers' accuracy while searching billion-scale code libraries fast enough to be practical, helping flag license and attribution risks.LINK

Autonomous AI research ran nearly 1,800 experiments on its own to design better AI architectures, discovering 105 top-performing designs, with its best beating a leading model by three times the gain of a rival.LINK

Mobile app agents get a smarter planning tool that predicts what happens next on a phone screen in plain words instead of images, boosting how often these assistants complete tasks successfully.LINK

See you tomorrow for a new dose of ☕️ Techpresso!