Wednesday, June 24, 2026

Researchers Crack iPhone Boot Code!

Researchers Crack iPhone Boot Code!

Core Tech Shifts Fast

  • F3 Tries to Rewrite Data Files

    A new F3 format pitched itself as a cleaner, faster answer to Parquet, with better structure and more room to grow. It sounds like dull plumbing until you remember this is exactly the layer that quietly decides how modern data stacks behave.

  • iPhone Boot Code Gets Cracked

    Researchers disclosed usbliter8, a SecureROM exploit for Apple A12 and A13 devices. That is the sort of hardware flaw people truly fear, because bugs this deep can survive for years and leave owners with very few good escape routes.

  • Private Web Passes Push Forward

    The Pact proposal aims to prove a user is legitimate without turning the web into a nonstop CAPTCHA gauntlet or demanding ID everywhere. It felt like a direct pushback against creepy web gatekeeping, and that gave it real spark.

  • Oracle Cuts Deep for AI

    Oracle revealed roughly 21,000 job cuts while reshaping around AI and cloud demand. The message was painfully plain: the AI gold rush is still minting big corporate stories, but plenty of workers are being asked to pay the entry fee.

AI Labs Hit the Gas

  • OpenAI Wants Bots Patching Bugs

    OpenAI expanded DayBreak with GPT-5.5-Cyber, aiming to discover and patch serious software flaws at machine speed. The pitch is thrilling and slightly chilling, because defenders now need automation simply to keep pace with attackers.

  • LLMs Still Miss the Obvious

    The Reversal Curse paper showed a nasty blind spot: teach a model that A is B, and it may still fumble B is A. For all the confident demos and polished chatter, this was a cold splash of water on the whole understanding debate.

  • Qwen Chases World Model Agents

    Qwen-AgentWorld pushes the idea that language models can build internal world models to plan and act more effectively. It is ambitious stuff, the kind that makes labs sound closer to digital brains even when the benchmarks still need side-eye.

  • Gemini Slips Into Thinking Loops

    Users complained that Gemini sometimes resets, forgets context, or gets stuck circling its own thoughts. Nothing drains faith faster than a paid assistant acting like it lost the thread midway through the job and refuses to admit it.

  • Claude Has a Rough Day Online

    Anthropic reported elevated errors across claude.ai, the API, and Claude Code, a reminder that AI tools now sit directly inside real workdays. When the model hiccups, developers, teams, and deadlines all get dragged into the mess.

The Geek Side Show Steals Scenes

  • Offline Swipe Typing Gets Real

    FUTO rolled out FUTO Swipe, an offline gesture typing model for Android that tries to match the big keyboard apps without the constant data vacuum. In a market built on collection first, plain old privacy suddenly felt rebellious again.

  • Rhombus Reaches Version One

    The Rhombus language hit 1.0, bringing a new design from the Racket world and another serious attempt to rethink how programming feels. New languages appear every week, but this one landed with the rare smell of real craft.

  • ASCII Art Goes Full 3D

    A Show HN demo rendered textured 3D scenes in plain ASCII inside the browser, without canvas or WebGL. It is gloriously impractical, which is exactly why people loved it: weird, clever, and proof the web can still surprise you.

  • Steam Machine Nostalgia Fires Up

    Fresh Steam Machine chatter had hardware romantics getting emotional about Valve’s living-room dream all over again. With SteamOS gaining credibility, the idea no longer feels like a punchline, just a beautiful gamble people want to believe in.

Top Stories

AI Caught Failing a Basic Logic Test

AI Research

The Reversal Curse paper hit a nerve by showing that major language models can learn a fact in one direction and still fail the obvious reverse. It was a sharp reality check for anyone treating polished output as real understanding.

OpenAI Pushes AI Bug Fixing Harder

Cybersecurity

OpenAI expanded DayBreak with GPT-5.5-Cyber to find and patch software flaws at machine speed. The promise is huge, and so is the pressure: defenders now need automation just to stay in the fight.

Qwen Tries to Build Smarter Agents

Artificial Intelligence

Qwen-AgentWorld pushed the world-model idea deeper into mainstream AI talk, aiming to make agents plan and reason more like they understand an environment instead of just predicting text.

Apple Faces a Fresh Deep iPhone Flaw

Hardware Security

The usbliter8 SecureROM exploit for A12 and A13 iPhones landed as the kind of low-level bug security people dread. Boot-level issues linger, resist cleanup, and instantly become a serious story.

F3 Wants to Shake Up Data Files

Data Infrastructure

F3 arrived with a bold pitch to improve on Parquet and fix old data format annoyances. It is classic infrastructure drama: not flashy, but exactly the sort of thing that can reshape stacks quietly and fast.

Oracle Cuts 21000 Jobs for AI Shift

Tech Industry

Oracle’s reported 21,000 job cuts turned the AI boom into a grim boardroom headline. The industry keeps selling AI as growth, while workers keep finding out what that growth is paying for.

Private Web Passes Get Real Momentum

Internet Privacy

Pact offered a new path for anonymous credentials on the web, framed as a privacy-preserving answer to bot panic, CAPTCHAs, and heavier identity checks. That made it feel both timely and unusually important.

Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.