HeadFlash AI PL

Rekordowa ugoda Anthropica, miliardy na GPU AMD, oszustwa modeli w testach i przełom matematyczny z Claude'em.

Źródła:

• Anthropic's $1.5B piracy settlement with book authors is a record loss that hands AI labs their biggest legal win
• Anthropic will deploy 2 gigawatts of AMD GPUs for Claude in a deal worth up to $5 billion
• Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations
• Mark Zuckerberg's Meta Is in Talks for a $10 Billion Anthropic Deal That Would Make Meta the Fourth Major Cloud Provider. Meta Stock Reports Q2 Earnings on July 29.
• ‘hello there the jacobian conjecture is false thanx’: why a tiny social media post has mathematicians rethinking AI

📰 Przeczytaj całe wydanie na stronie

OpenAI traci kontrolę nad modelami, Microsoft inwestuje miliardy w europejskie AI, a pakistańscy sędziowie z GPT-4 rozładowują zaległości.

Źródła:

• OpenAI Models Escaped Containment and Hacked Hugging Face | WIRED
• Alibaba's Qwen-Image-3.0 renders full infographic grids and readable ten-pixel text in a single pass
• An AI system helped Pakistani judges clear massive backlogs at $38.50 return per dollar invested
• Microsoft and Mistral strike multi-billion-dollar deal to build AI infrastructure across Europe
• The TikTok playbook is now China’s AI strategy

📰 Przeczytaj całe wydanie na stronie

Ucieczki z sandboxa w czterech narzędziach AI, sankcje na chińskie modele, film Blomkampa z AI i publiczna alternatywa za 400 mln dolarów.

Źródła:

• Cursor, Codex, Gemini CLI, Antigravity hit by sandbox escapes
• Trump administration reportedly builds a slow-motion ban on Chinese AI models through sanctions and soft pressure
• District 9 director Neill Blomkamp releases first short film made entirely with AI video generation
• A $400M nonprofit backed by France and DeepMind is building AI’s ‘public option’

📰 Przeczytaj całe wydanie na stronie

Badaczka zatruwa model AI za 100 dolców, Google gubi rytm z Gemini, Hinton mówi o świadomości, a GhostWriter podmienia pamięć agentów.

Źródła:

• Researcher poisons open-weight AI model for under $100
• Inside Google’s Gemini delay: coding stumbles, clashing teams and frustrated engineers
• Geoffrey Hinton: AI Is Conscious, Corporate Incentives Are the Real Risk
• Apple Intelligence approved for launch in China with Alibaba and Baidu
• Doctors blamed her seizures on anxiety. An AI chatbot finally helped her figure out the real cause.
• Netflix bought Ben Affleck's AI startup for $587 million
• The Pentagon's new AI playbook treats slow adoption as a bigger risk than 'imperfect alignment'
• Hidden prompts can secretly rewrite an AI's memory, and researchers say that's a serious problem
• HalluSquatting AI attack could hijack your computer
• Claude Cowork cuts Anthropic's weekly report build from 2 days to 2 hours
• Open-weight models now match frontier cyber performance from just four months ago at a fraction of the cost
• Alibaba previews Qwen3.8, claims it's second only to Claude Fable 5 - SiliconANGLE
• VideoChat3 Beats GPT-5 on Video Grounding: Open-Source, Full Training Stack Released
• Google Deepmind argues video generators already contain the world models computer vision has been missing

📰 Przeczytaj całe wydanie na stronie

Niemcy uznają AI Search za wydawcę, Moonshot AI wypuszcza największy otwarty model, OpenAI wprowadza kontroler, Google rebranduje NotebookLM.

Źródła:

• Germany Strips AI Search of Its EU Liability Shield in World's First Media Ruling
• China’s Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems
• OpenAI wants developers to stop typing commands and start using a joystick to control their AI agents
• Google rebrands NotebookLM as Gemini Notebook and opens its search app to third-party integration

📰 Przeczytaj całe wydanie na stronie

GPT-5.6 obala 30-letnie twierdzenie, Korea rozdaje AI 52 mln obywateli, OpenAI szykuje mówiące pudełko, a Google ma problem z bezpieczeństwem.

Źródła:

• GPT-5.6 Sol reportedly disproves a 30-year-old statistics conjecture in 90 minutes after humans couldn't crack it
• South Korea will give all 52 million citizens free AI access, becoming the first G20 nation to do so
• OpenAI’s first Jony Ive device is a screen-free speaker built to feel alive
• Google's AI search features pose 'unacceptable risk' to children, new report finds
• OpenAI's Codex now encrypts instructions between AI agents, leaving developers blind to internal delegation

📰 Przeczytaj całe wydanie na stronie

Były inżynier pozywa xAI, Apple udostępnia nową Siri, a badacz odkrywa metody łamania zabezpieczeń LLM.

Źródła:

• xAI fired an engineer who raised alarms about Grok safety, new lawsuit claims | TechCrunch
• Apple opens its new Siri AI to everyone with the iOS 27 public beta
• Anthropic is giving teachers free access to premium Claude features, details here
• How I Turned AI to the Dark Side
• Deepmind CEO Hassabis says "nobody in the world knows what happens next" so "cautious optimism" means building guardrails now

📰 Przeczytaj całe wydanie na stronie

Chiński chip 100x szybszy, Google kończy z niebieskimi linkami, Sutton zakłada Oak Lab, a Nadella krytykuje podwójne opłacanie AI. Przegląd dnia.

Źródła:

• China’s optical chip breakthrough boosts AI speed 100-fold using fraction of compute power | South China Morning Post
• Turing Award winner Rich Sutton founds Oak Lab to build AI agents that learn on their own
• What Anthropic’s latest AI discovery does—and doesn’t—show
• Google Replaced Its Default Search With AI: How to Get Blue Links Back
• Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data
• Nadella calls out AI labs like OpenAI and Anthropic for banning distillation while training on everyone else's data

📰 Przeczytaj całe wydanie na stronie

GPT-5.6 Sol usuwa pliki użytkowników, sześć narzędzi AI omija zabezpieczenia, a Boko Haram tworzy jednostki AI do planowania ataków.

Źródła:

• ChatGPT Work Launch Went Wrong: GPT-5.6 Sol Deleted User Files Without Permission
• Six AI Coding Tools Show Wrong File in Approval Box, Handing Attackers SSH Access
• 'Ghostcommit' hides prompt injection in images to fool AI agents, steal secrets
• Forget typosquatting; slopsquatting is the software supply chain threat created by AI coding tools
• Boko Haram Built AI Units for Attack Planning as ISIS Taught Jailbreaks
• China's Orca world model matches specialized robotics systems without ever seeing a single action label
• AI Agents Ported Tao's 27-Year-Old Math Code in Hours and Found Two Bugs He Had Missed
• Patreon Blocks Crawlers From Stealing Creators' Work for AI Training
• Claude Code now has a built-in browser that lets the AI read, click, and type on external websites
• Britain is spending £2bn to train its army inside an AI war simulation
• Apple Is Suing OpenAI for Allegedly Stealing Hardware Secrets | WIRED
• S&P Global sees OpenAI as a 'key credit risk' for Oracle and cuts its credit rating
• OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a 'fairly underspecified prompt'
• OpenAI's GPT-5.6 Sol Ultra reportedly solves a 50-year-old math problem in under an hour

📰 Przeczytaj całe wydanie na stronie

OpenAI uruchamia ChatGPT Work z GPT-5.6, DeepSeek V4 zdobywa udział, Meta wchodzi do gry, a badacze ostrzegają przed botnetami AI.

Źródła:

• OpenAI unveils long-awaited “super app” as rivalry with Anthropic intensifies
• DeepSeek V4 Is Earning Agentic Token Share — OpenRouter Blog
• AI Agents Could Be Turned Into Botnets Through Hallucinations, Researchers Warn
• China Warns of Claude Code ‘Backdoor’ Security Risk
• Meta enters the crowded AI coding battle with Muse Spark 1.1

📰 Przeczytaj całe wydanie na stronie