☕️ Millions of iPhones can be hacked with new tool

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

📱 Millions of iPhones can be hacked with new tool

🚀 OpenAI launches GPT-5.4 Mini and Nano models

🛡️ DOD calls Anthropic an unacceptable national security risk

🍎 Apple cracks down on "vibe coding" apps

🎵 Judge rules Apple can remove apps “with or without cause”

🥽 Meta to discontinue key metaverse product for VR headsets

Plus: 🎁 13 other news you might like, 🧰 5 tools, and 📚 3 papers.

📱 Millions of iPhones can be hacked with new tool LINK

A spyware tool called "Darksword" that can hack hundreds of millions of iPhones was found planted on dozens of Ukrainian websites, according to researchers from Lookout, iVerify, and Google.

Google observed commercial vendors and suspected state-linked hackers using Darksword in campaigns against targets in Saudi Arabia, Turkey, Malaysia, and Ukraine, with some tied to Turkish firm PARS Defense.

Apple has patched the underlying bugs, but an estimated 220 million to 270 million iPhones still run exposed iOS versions because many people do not install updates, researchers said.

🚀 OpenAI launches GPT-5.4 Mini and Nano models LINK

OpenAI has released GPT-5.4 mini and GPT-5.4 nano, two smaller and cheaper models designed for fast, high-volume AI workloads like coding assistants, subagents, and multimodal applications that process images in real-time.

GPT-5.4 mini runs more than twice as fast as GPT-5 mini and scores 54.38% on SWE-bench Pro compared to 45.69%, approaching GPT-5.4-level pass rates while costing roughly one-third as much.

GPT-5.4 nano is API-only at $0.20 per million input tokens, aimed at classification, extraction, and ranking tasks, while mini is available in API, Codex, and ChatGPT across all user tiers.

🛡️ DOD calls Anthropic an unacceptable national security risk LINK

The U.S. Department of Defense has formally called Anthropic an "unacceptable risk to national security" in a court filing, pushing back for the first time against the AI company's lawsuits over its supply chain label.

The DOD argues in a 40-page filing that Anthropic might disable its technology or change its model's behavior during warfighting operations if the company believes its corporate "red lines" are crossed.

Anthropic had signed a $200 million Pentagon contract but later said it did not want its AI used for mass surveillance of Americans or in lethal weapons targeting and firing decisions.

🍎 Apple cracks down on "vibe coding" apps LINK

Apple is pushing back on "vibe coding" apps that let people build software by typing text prompts into an AI system, telling some developers their apps violate existing App Store rules.

Apple cites App Store Guideline 2.5.2, which says apps cannot download, install, or execute code that introduces or changes features or functionality after passing through the review process.

A possible fix for at least one affected app is generating previews in a browser instead of inside the vibe coding app itself, which could satisfy Apple's existing rules.

🎵 Judge rules Apple can remove apps “with or without cause” LINK

A federal judge ruled that Apple's developer agreement lets it remove apps "with or without cause," dismissing a lawsuit from free music streaming app Musi that sought to return to the App Store.

Musi streamed music from YouTube without striking its own deals with copyright holders, displayed its own ads, and was removed from the App Store in September 2024 after YouTube raised intellectual property claims.

The judge also sanctioned Musi's lawyers for "mak[ing] up facts to fill the perceived gaps in Musi's case," and dismissed the lawsuit with prejudice, meaning Musi cannot refile the same claims.

🥽 Meta to discontinue key metaverse product for VR headsets LINK

Meta is shutting down Horizon Worlds on its Quest VR headsets, meaning its flagship metaverse platform will no longer be available on its flagship metaverse-experiencing device and will become mobile-only.

Users who want to keep accessing Horizon Worlds on a headset must download it before March 31, and the app will stop working entirely on Quest devices after June 15.

The move follows apparent budget cuts at Meta's reality labs division last December, which raised concerns about the future of Horizon Worlds and the broader metaverse effort.

Other news you might like

• Nvidia Says It Is Restarting Production of AI Chips for Sale in ChinaLINK

• Jensen Huang says gamers are 'completely wrong' about DLSS 5 — Nvidia CEO responds to DLSS 5 backlashLINK

• Microsoft shakes up Copilot AI leadership team, freeing up Suleyman to build new modelsLINK

• Mistral bets on ‘build-your-own AI’ as it takes on OpenAI, Anthropic in the enterpriseLINK

• Apple Home hardware chief leaves for OuraLINK

• Microsoft weighs legal action over $50 billion Amazon-OpenAI cloud deal, FT reportsLINK

• How World ID wants to put a unique human identity on every AI agentLINK

• The Pentagon is planning for AI companies to train on classified data, defense official saysLINK

• Meta’s Manus launches ‘My Computer’ to turn your Mac into an AI agentLINK

• US securities regulator issues long-awaited crypto guidanceLINK

• China Ramps Up Scrutiny of Meta’s Acquisition of ManusLINK

🧰 Trending tools

Lightfield: an AI-powered CRM that automatically populates from emails, meetings, and calls, eliminating manual data entry and answering pipeline questions in plain English.LINK

Genie by Databox: an AI-powered business intelligence platform that simplifies data analysis by connecting multiple sources, creating custom metrics, and generating automated insights without complex setup.LINK

GPT-5.4 mini and nano: compact AI models optimized for faster inference and lower costs while maintaining strong performance for common development tasks.LINK

OpenObserve: open-source observability platform for logs, metrics, and traces that reduces storage costs by 140x compared to Elasticsearch with quick deployment.LINK

NetLab: a browser-based network simulation tool for practicing network configurations and topology design, serving as a Cisco Packet Tracer alternative.LINK

📚 Trending papers & reports

Current AI systems lack true autonomous learning because they require humans to define goals, select training data, and engineer rewards, unlike biological systems that self-generate learning objectives through intrinsic motivation.LINK

Painted umbrella patterns fooled commercial drone vision systems into crashing 100% of the time by exploiting how AI interprets visual obstacles, costing under $50 to execute.LINK

Smart glasses usage patterns succeed for quick tasks like timers and reminders but fail for complex queries, with users abandoning voice commands 40% of the time due to social awkwardness and unreliable responses.LINK

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

See you tomorrow for a new dose of ☕️ Techpresso!