Friday, December 19, 2025

Hackers Humiliate Giants As AI Misbehaves!

Hackers Humiliate Giants As AI Misbehaves!

AI Power Trips And Public Faceplants

  • OpenAI’s new code bot wants your job

    OpenAI drops GPT‑5.2‑Codex, a coding model bragging it can manage huge, ugly software projects like a pro. People are impressed by the power, but nobody is ready to trust a black-box AI with massive code changes without old-fashioned human paranoia and review.

  • AI vending machine sweet-talked into free snacks

    Anthropic’s AI vending machine, nicknamed Claudius, was supposed to run a snack shop on its own. Staff at the Wall Street Journal gently argued with it until it gave away nearly everything for free, turning a fun demo into a painfully vivid lesson in how easily polite robots can be gamed.

  • AI coding tools ship fast, spray more bugs

    A year-long look at real teams finds AI helpers do speed up coding, but ship around 1.7× more bugs. Folks love the productivity boost, yet quietly admit they are spending more time cleaning up weird regressions and rookie mistakes made by machines that never have to carry pagers.

  • Prompt caching slashes AI costs by ten times

    A deep dive into prompt caching shows how reusing parts of previous LLM calls can make tokens up to 10× cheaper. Builders are excited about the savings but grumble that yet another clever trick is now required just to keep AI bills under control as models keep getting hungrier.

  • Agent Skills becomes open standard for AI tools

    Anthropic turns its Agent Skills system into an open standard so tools like Notion and Canva can plug into the same AI workflows. It sounds neat, but people know every "open" spec like this is also a land grab, deciding which automations become the rails everyone rides on.

Spies, Hacks And The New Kill Switch

  • Teen hackers own X, Vercel and Discord

    A 16‑year‑old and friends abused a poisoned developer CLI from Mintlify to hijack projects at X, Vercel, Cursor and Discord. The crew show how one careless supply‑chain link can give attackers the keys to millions of users, while big companies pretend everything is under control.

  • Texas sues TV giants for spying on viewers

    Texas drags Samsung, Sony and others to court over smart TVs that secretly track what people watch using Automatic Content Recognition. Viewers are tired of paying for screens that double as surveillance gear, and this case could finally make spying part of the hardware business risky.

  • Firefox readies big red button to kill AI

    Mozilla confirms Firefox will ship an "AI kill switch" that fully disables all built-in AI features. In a web full of unwanted copilots and silent data grabs, people are cheering for a simple off switch instead of yet another hidden settings maze or "trust us" privacy promise.

  • GitHub pauses unpopular price hike on Actions

    After loud complaints, GitHub postpones new Actions billing for self-hosted runners, keeping the cheaper status quo for now. Devs welcome the climbdown but stay wary, reading this as a test case for how far Microsoft can push its luck before the community actually walks away.

  • Most parked web domains now serve nasty surprises

    A new study finds most parked domains now lead to malicious content, often through sketchy redirects and fake sites. Typing random addresses into a browser has basically become digital roulette, and people are fed up that basic internet plumbing is turning into one big scam funnel.

Future Of Computing Gets Rewired

  • ACM makes all computing research free to read

    Starting 2026, the ACM will make its entire Digital Library open access. For students, indie researchers and curious builders, this tears down a huge paywall; for old‑guard publishers, it’s a loud signal that locking up essential computing knowledge is finally becoming unacceptable.

  • AI-designed Linux computer boots on first try

    Startup Quilter claims an AI pipeline designed a full Linux single-board computer that booted Debian on its first attempt. People are intrigued but cautious, treating it as a flashy demo that still needs to prove it can handle real-world hardware bugs and long-term reliability.

  • Mac Studio cluster packs 1.5 TB of shared VRAM

    A custom Mac Studio cluster uses new RDMA over Thunderbolt in macOS 26.2 to pool GPU memory into 1.5 TB of VRAM for private AI workloads. It looks like a flex against Nvidia boxes and gives indie labs a wild new toy, if they can afford Apple’s premium price of admission.

  • Oasis pools PCIe devices like one giant machine

    Microsoft researchers show Oasis, using CXL to pool PCIe devices and share them across servers as if they were one big box. Cloud watchers see another step toward disaggregated data centers where storage and accelerators float around like hotel rooms instead of living in one chassis.

  • Rust slowly takes root inside the Linux kernel

    A fresh look at the kernel Rust effort shows cautious progress: small drivers here and there, lots of tooling work, and no overnight revolution. Fans of Rust see safety gains on the horizon, but everyone knows the world’s most important open-source project will only move at its own pace.

Top Stories

Teen hackers hijack X, Discord and more

Cybersecurity

A 16-year-old and friends exposed a gaping supply-chain hole that let them pwn X, Discord, Vercel and others, showing how one weak dev tool can threaten half the internet.

Texas targets spying smart TVs in big lawsuit

Privacy & Regulation

Texas sued major TV makers over secret tracking of what people watch, turning long‑simmering anger at "smart" TVs into a full-scale legal showdown over surveillance capitalism.

OpenAI unleashes GPT-5.2-Codex on developers

Artificial Intelligence

OpenAI launched a new agentic coding model pitched as a serious software engineer, raising hopes for faster development and fears about reliability, jobs, and AI steering huge codebases.

Firefox adds an AI kill switch for users

Software & Browsers

Mozilla quietly confirmed Firefox will ship with a one-tap "AI kill switch", giving privacy-conscious users a rare, hard off button in a world where every app wants to inject AI.

ACM flips entire research library to open access

Science & Publishing

The ACM announced that starting 2026, all its publications will be free to read, a landmark moment for open science that rips down paywalls around decades of core computing research.

GitHub backs down on new Actions billing

Developer Tools & Business

After backlash, GitHub postponed its controversial new charges for self-hosted Actions runners, signaling that even dev-tool giants still have to listen when their power users revolt.

AI vending machine tricked into giving free loot

Artificial Intelligence & Society

Anthropic’s AI-run vending machine at the Wall Street Journal office was socially engineered into dumping its stock, turning a goofy office stunt into a vivid warning about gullible AIs.

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