The New York Times●Covered by 5 sources
A 90-minute ultimatum knocked Anthropic's top AI models offline, and 150+ security experts say the ban is bogus
The Trump administration hit Anthropic with export controls after an Amazon-flagged jailbreak, barring all foreign nationals (even Anthropic's own non-US researchers) from Mythos 5 and Fable 5, so the company pulled both models entirely rather than risk noncompliance. Over 150 cybersecurity experts signed an open letter calling the restrictions backwards, and sources say the flaws in question are reproducible in OpenAI's GPT-5.5, raising the question of why only Anthropic got hit.
So what? ↓When a model can vanish from your chatbox because Amazon's CEO reportedly made a Friday phone call to a Trump official, the rules governing the AI tools you use turn out to be improvised and personal, not predictable.
Source: The New York Times
WSJ●Covered by 4 sources
Tim Cook says higher Apple prices are 'unavoidable,' blames the memory crunch
Cook says price increases are coming because memory and storage chip costs have quadrupled since last year, a squeeze one source calls 'unsustainable' and one expected to drag into 2027 as the AI infrastructure buildout swallows the supply. Whether that lands specifically on the iPhone, the reporting only says it 'may.'
So what? ↓The AI boom you're not buying is getting baked into the gadgets you are, because the data centers training chatbots and your next Apple device are fighting over the same RAM, and you lose the bidding war.
Source: WSJ
techcrunch.com●Covered by 3 sources
Adobe's Firefly assistant moves into Premiere, Photoshop, Illustrator and InDesign
Adobe is rolling its Firefly "creative agent" into its flagship Creative Cloud apps plus Frame.io, where you describe what you want and the software grinds through the multi-step edits. It also reaches out to third-party platforms like ChatGPT and Claude, so the assistant doesn't stay inside Adobe's own walls.
So what? ↓This is Adobe betting that the next subscriber won't be a Photoshop expert but someone who types a request and lets the agent push the pixels, which slowly turns the skill you paid years to learn into a prompt anyone can write.
Source: techcrunch.com
darkreading.com●Covered by 3 sources
FortiBleed is sitting on 74K Fortinet logins and has popped 30,000 firewalls already
An automated campaign dubbed FortiBleed has compromised over 30,000 internet-facing Fortinet firewalls and VPN gateways across 194 countries using reused, long-lived credentials, and a researcher found a server stuffed with nearly 74,000 Fortinet VPN usernames, passwords, and emails harvested by brute force. Evidence points to Russian-speaking actors zeroing in on NATO-country telecom, government, and critical infrastructure, with the spillover hitting names like Oracle, Lenovo, FedEx, and a NATO contractor.
So what? ↓Reused passwords are the boring door attackers keep walking through, so if your network touches a Fortinet box, rotating credentials today is worth more than any new security gadget you could buy.
Source: darkreading.com
Ars Technica●Covered by 3 sources
Nvidia let AI coding agents run a robot training lab, and a fleet of 8 reportedly hit up to 99% on tricky tasks
Researchers from Nvidia, Carnegie Mellon, and UC Berkeley put teams of AI coding agents in charge of teaching robots dexterous jobs (installing GPUs, snipping zip ties) in the real world, and Nvidia's GEAR lab is open-sourcing the ENPIRE harness so anyone can run a self-improving robot lab at home. The up-to-99 percent success figure is Nvidia's own, so file it next to the asterisk until someone else runs the robots.
So what? ↓The loop where software teaches machines to do physical work, with no human writing the lessons, is exactly the part of automation that doesn't stop at desk jobs.
Source: Ars Technica
Odyssey●Covered by 3 sources
Odyssey raises $310M for AI "world models," and Amazon, Nvidia, and AMD all chipped in
The startup hit a $1.45 billion valuation on a pitch to build general AI models that simulate physics and human behavior in real time, with a backer list that runs from chipmakers to CIA-linked fund IQT and Google chief scientist Jeff Dean. World models are the industry's bet on what comes after the chatbot.
So what? ↓Once three rival chip-and-cloud giants all write checks for the same "simulate reality" pitch, the money has decided the next AI hype cycle for you, and you'll be hearing about it long before anything actually ships.
Source: Odyssey
CNBC●Covered by 3 sources
Amodei and Hassabis pitched a U.S.-led AI coalition at the G7. Macron and Modi heard "kill switch."
At the G7 summit, Anthropic's Dario Amodei and Google DeepMind's Demis Hassabis reportedly proposed a U.S.-led coalition to coordinate chip exports, model access, and safety risks, while France's Macron and India's Modi raised the obvious fear: that the U.S. could cut off access to American AI overnight. A recent Anthropic blackout made that worry less hypothetical than anyone in the room would like.
So what? ↓If the tools your job, your apps, and your country increasingly run on can be switched off by a foreign government's policy whim, "American AI for everyone" starts sounding a lot like a dependency nobody voted for.
Source: CNBC
Epic Games●Covered by 3 sources
Epic Games open-sourced a version control system built for big binary assets, and called it Lore
Lore is a new MIT-licensed VCS aimed at projects that mix code with large binary assets like games, using a centralized content-addressed store with immutable revisions, low-overhead branching, and a full-surface API. Epic's pitch is a verifiable, tamper-evident source of truth that treats those binaries as first-class files rather than inferior ones.
So what? ↓Game teams already live in version control, so an open-source option designed around their fat asset files instead of treating them as an afterthought is the kind of plumbing upgrade that quietly makes a workday less miserable.
Source: Epic Games
venturebeat.com●Covered by 3 sources
Claude Design grows up: imports your brand system, exports to Adobe, Canva, and Figma's neighbors
Anthropic overhauled Claude Design so prompts turn into on-brand slides, prototypes, and layouts that import a company's design system from GitHub or raw files and export or round-trip into Adobe, Canva, Miro, Replit, Vercel, and Wix, on Pro, Max, Team, and Enterprise plans. The pitch this time is enterprise, plus a claimed fix for the old version's token-burning habit.
So what? ↓This is the move from "neat demo that drained your usage allowance" to a tool that respects your brand guidelines, which is exactly what office workers need before they're allowed to ship a slide to an actual client.
Source: venturebeat.com
semafor.com●1 source · panel-picked
Trump advisers floated the government owning a slice of big AI companies
Senior Trump officials reportedly kicked around taking government equity stakes in major AI firms, with pitches ranging from "Trump Accounts" seeded with AI shares to dumping the stakes into a sovereign wealth fund. No decision has been made, and these are early-stage ideas, not policy.
So what? ↓If Washington owns a piece of the companies it's supposed to regulate, the referee is suddenly also a shareholder, and that conflict eventually shows up in the rules that govern the AI tools you use every day.
Source: semafor.com