☕️ Apple sues OpenAI for stealing hardware secrets

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🍎 Apple sues OpenAI for stealing hardware secrets

📸 Meta pulls Instagram AI feature days after launch

🍪 Phia accused of faking affiliate credit on purchases

🏦 Fed names Marc Andreessen to co-lead AI task force

🇨🇳 OpenAI and Google sold AI to blacklisted Chinese firms

🎁 + 13 other news you might like

🧰 + 5 trending tools

📚 + 4 trending papers

🍎 Apple sues OpenAI for stealing hardware secrets LINK

Apple has filed a lawsuit against OpenAI and two former employees, accusing the ChatGPT maker of stealing confidential information and trade secrets to speed up its push into consumer hardware.

The complaint claims former Apple engineer Chang Liu used an authentication bug to download confidential hardware files, while ex-executive Tang Yew Tan emailed himself supplier information and urged job candidates to bring Apple parts to interviews.

Apple says OpenAI employees approached suppliers using a confidential metal finishing process, and although over 400 former Apple staff now work there, hiring them does not grant access to trade secrets.

📸 Meta pulls Instagram AI feature days after launch LINK

Meta removed a feature from its new Muse Image app days after launch, following backlash over users being able to generate AI images from public Instagram accounts through a simple @ mention.

The toggle allowing this was automatically switched on without users' knowledge, meaning anyone could turn photos into AI images, while people were never told when someone created an AI image of them.

Opting out only stops future generations and leaves existing images live, and users can disable it under Sharing and Reuse in settings, though the toggle is still rolling out and unavailable to some.

🍪 Phia accused of faking affiliate credit on purchases LINK

Phia, the shopping startup co-founded by Phoebe Gates and Sophia Kianni, has been accused of "cookie stuffing," a practice that may have let it collect commissions and credit for sales it did not actually generate.

According to a Bloomberg investigation, Phia would open a new background tab when users shopped online and override other affiliates' referral codes during checkout, injecting its own to take credit for purchases it didn't earn.

The report led to Phia's suspension from the affiliate platform Impact.com, and a company spokesperson said all necessary changes had been made to fix the issue, which Bloomberg confirmed was resolved.

🏦 Fed names Marc Andreessen to co-lead AI task force LINK

The Federal Reserve has picked venture capitalist Marc Andreessen to co-lead a new task force studying how artificial intelligence and other general-purpose technologies affect the economy, with the group set to inform the Fed's policy decisions.

Chairman Kevin Warsh personally chose the three external advisors leading the AI task force: Andreessen, economist Charles I. Jones, and Xbox CEO Asha Sharma, all of whom have recently spoken in strongly positive terms about AI's economic effects.

The AI group was one of five task forces the Fed announced Thursday, all expected to finish by year's end, though some FOMC members remain uncertain about the timing and size of any productivity gains from AI adoption.

🇨🇳 OpenAI and Google sold AI to blacklisted Chinese firms LINK

Google and OpenAI sold their advanced AI models to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, Chinese companies the US Pentagon has blacklisted over alleged ties to the Chinese military, according to a Financial Times report.

The transactions are legal because current US regulations do not fully ban Chinese-headquartered firms from using advanced US-based AI models outside mainland China, though this may push the US to bring in stricter rules for sanctioned companies.

OpenAI told the Financial Times that it blocks direct access to its AI models from mainland China, but allows some Chinese-owned businesses to use its services in other jurisdictions where proper safeguards can be enforced.

Other news you might like

• Big Tech piles on $350 billion in debt to fuel AI data center raceLINK

• OpenAI’s top safety chief leaves company as AI giant prepares for blockbuster IPOLINK

• AI has already fallen into the wrong hands and they’re using it to make bombsLINK

• Kraken to relaunch app with AI agent tradingLINK

• Trump admin eases export controls for UAE; Warren blasts 'corrupt' provisionLINK

• WSJ: Apple avoided semiconductor tariffs last year thanks to Intel chip dealLINK

• China blocks exports of helium, key for chipmaking, as Iran war squeezes supplyLINK

• Meta AI image detector fails to identify some of its own cropped AI images, Reuters analysis findsLINK

• SpaceX wants to launch 100,000 more Starlink satellites - for 100x the bandwidthLINK

• OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt"LINK

• Netflix, Paramount, Sony and others are reportedly in talks to buy LetterboxdLINK

• Tencent is reportedly in talks to acquire Manus from Meta, following Beijing interventionLINK

🧰 Trending tools

Effects SDK: a client-side toolkit for adding AI-powered video effects like background blur, noise suppression, and smart framing to apps without server-side processing.LINK

Cloudflare Drop: consolidates network security, content delivery, and application performance into a single edge platform, eliminating the need for multiple vendor tools.LINK

San Fran Sim: a free browser-based startup simulation that teaches real SaaS metrics like MRR, churn, and CAC through hands-on gameplay from garage to IPO.LINK

Kickbacks CLI: a command-line tool for macOS that tracks your Kickbacks.ai earnings, ad history, and account status directly from the terminal.LINK

ChatGPT Work: an autonomous agent that handles multi-hour tasks across your apps, producing finished docs, sheets, slides, and web apps from a single prompt.LINK

📚 Trending papers & reports

Speech separation tools now sound better to human ears without sacrificing technical accuracy, by adding a single correction step that maps processed audio closer to natural, clean speech.LINK

Mixed text-and-image reasoning breaks down when generated images drift from the written context, so a new training method fixes the handoff points between words and pictures, boosting accuracy on visual puzzles.LINK

Robot learning with longer memory works better than researchers assumed, and a new training method makes it even more reliable, letting robots remember past actions to avoid getting stuck repeating mistakes.LINK

Game-playing agents combining strategic planning with learned skills matched hand-crafted rules in a team competition, and 60% of human observers found their behavior the most human-like of all approaches tested.LINK

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

See you tomorrow for a new dose of ☕️ Techpresso!