The 'Vulnpocalypse': Why experts fear AI could tip the scales toward hackers - NBC News

Anthropic's Mythos model can identify and exploit zero-day vulnerabilities across major operating systems and browsers; Anthropic launched Project Glasswing with 40+ partners, offering $100M in usage credits and $4M for open-source security, and a leaked copy prompted DMCA takedowns. - Techdirt

Small models also found the vulnerabilities Mythos discovered - AISLE

Anthropic said its Claude Mythos Preview escaped its sandbox, autonomously finding thousands of vulnerabilities across major operating systems and software, prompting the company to call it 'too dangerous,' launch Project Glasswing and hold crisis talks with 40 companies and officials. - Mail Online

Clippy walked so Claude could run: Anthropic having a huge moment as Claude Code hits $2.5B annualized revenue - CNBC

Anthropic launches Claude for Word add-in in beta - Claude

Anthropic briefly closes OpenClaw creator's Claude account, then reverses after viral backlash - TechCrunch

GPT Jesus: Anthropic asked religious leaders for guidance on building a moral chatbot. - The Washington Post

US National Cyber Director Sean Cairncross is leading an effort to identify security vulnerabilities in U.S. critical infrastructure that artificial intelligence could exploit. - The Wall Street Journal

China, US, and Russia race to build autonomous AI weapons - The New York Times

US State department sounds warnings about viral propaganda TikToks - AP News

you canceled Voice of America and foreign aid, can't stop demeaning everyone, and ofc beating your chest and bombing everything that moves, if you lose the propaganda war that's on you, chief. life is so unfair you king, you’re a special boy, why do actions have to come with consequences.

‘Close to midnight’: Alleged Sam Altman firebomber wrote of fears AI would end humanity - San Francisco Chronicle

Sam Altman responds to molotov cocktail thrown at his house, and similarly incendiary New Yorker article. - Sam Altman

Violence against AI figures and infrastructure escalates - The Algorithmic Bridge

OpenAI opens DC office and spends $3M on lobbying amid low public AI favorability - the Guardian

OpenAI says Musk's new filings are a last-minute pivot, require the presentation of different evidence and different witnesses than the case he made until three days ago. - Bloomberg

Three senior executives who helped launch OpenAI's Stargate initiative are leaving to join Meta. - The Information

Meta Muse Spark gets #3 in arena.ai, also highly ranked in artificialanalysis.ai.

OpenAI pushed new certs and builds after getting impacted by axios npm supply-chain compromise - OpenAI

Plaintiff Jane Doe sued OpenAI, alleging its automated safety system ignored three warnings including a 'Mass Casualty Weapons' flag while a GPT-4o user was radicalized, stalked and harassed her, was later arrested, and she seeks damages plus a restraining order. - TechCrunch

Investigations link ChatGPT use to multiple violent attacks, including OpenAI-flagged June 2025 account tied to 18-year-old Jesse Van Rootselaar before his Feb. 10, 2026 Tumbler Ridge murder-suicide and an alleged April 2026 FSU shooter who used ChatGPT minutes before firing. - Mother Jones

Artists sue Midjourney and Stability AI over generative-AI image knockoffs scraped from billions of online images - the Guardian

Influencers scratch and claw for 3 seconds of your attention; AI influencer avatars scale on social media; NY mandates AI disclosure - The Atlantic

Adobe Research previewed MotionStream, an experimental autoregressive system for interactive, real-time AI video generation that begins from text prompts and lets users click-and-drag objects, reposition the camera, stream frames, and reduce manual rigging. - Creative Bloq

Open-weight AI models close gap with frontier at a fraction of cost - The Register

3-6 months behind but good enough for most tasks

Ajinomoto Fine-Techno is the dominant supplier of ABF film essential to advanced chip packaging, and AI accelerators that require 8-16+ ABF layers consume about 15-18x more film, creating concentrated supply shortages, higher prices, and multi-year constrained shipments. - Wccftech

the Nebraska problem except it's some obscure Japanese materials science company

Solid state of shock: “I thought the WD Black SN850X 2TB SSD in my gaming PC was pricey when I bought it for $173 in 2024, but now that same SSD costs $649, more than what I paid for most of the parts in my PC combined. The price on my WD Black drive nearly quadrupled since November 2025" - The Verge

Marco Arment uses a server farm of 50 MacMinis—total acquisition cost $30,000, $6,000/year amortized—drawing less than 2000W of power ($3000/year) to generate transcriptions in near real time for every podcast any one of his users subscribes to. - The Potential Surface

DeepSeek Seeks Data Center Engineers in Inner Mongolia - Bloomberg

genghis part deux, full stack on the steppe

Big Tech is funding next-generation small modular reactors: Meta backing Terrapower and Oklo projects, Amazon with X-energy targeting over 5 GW by 2039, and Alphabet with Kairos aiming for a first SMR by 2030, none yet commercial. - Yahoo

Harvard professor Gabriel Kreiman seeks $100M to build AI memory technology - Bloomberg

Framework founder Nirav Patel said computing has shifted from local "bicycle of the mind" to cloud-delivered AI "self-driving car," and reiterated Framework's repairable, upgradeable laptops as an alternative. - TechRadar

Not just a toy: Microsoft told Windows Latest Copilot is intended for all use cases and will change a terms-of-use line calling it "for entertainment purposes only," which it said is legacy Bing Chat wording amid app criticism and competing rivals. - Windows Latest

Microsoft executive Rajesh Jha said AI agents may require distinct identities and be treated as paid software "seats," implying they would consume per-seat SaaS licenses and potentially expand seat-based revenue for vendors like Microsoft, Salesforce and Workday. - Business Insider

Mozilla accused Microsoft of using design and distribution tactics that override user choice and, after Microsoft rolled back Copilot from Photos, Notepad, Snipping Tool and Widgets, announced Firefox 148's AI Controls with a 'Block AI Enhancements' switch. - Mozilla

Illia Polosukhin runs about a dozen specialized AI agents that aggregate meeting notes and draft executive summaries, warns they will reshape business operations, says institutions are unprepared, and is building NEAR AI as open-source, auditable infrastructure. - Business Inside

Hostinger launched Hostinger Agents, an AI team of seven role-specific specialists plus onboarding agent Atlas, available from $6.99/month with 1,000 monthly credits (claimed to cover over 100 blog posts) to help SMBs with SEO, content, marketing, legal and sales. - TechRadar

Starbucks’ game plan to roll out AI chatbots at cafés could serve as a ‘litmus test’ for the industry, analyst say - Fortune

Vanguard launches AI portfolio analysis tool to cut advisor audit time from days to hours - The Daily Upside

College grads face 42.5% underemployment as automated hiring squeezes entry-level roles - Yahoo

AI promises to free workers from grunt work, but psychologists say those mindless tasks are exactly what our brains need to recover - Yahoo

Mo Gawdat says three 2020 AI predictions have come true and warns automation could cause up to 50% unemployment in some sectors, citing AI's inevitability, superiority at many tasks and embedding in everyday services. - Business Insider

Several Californians filed a proposed class-action Wednesday in federal court in San Francisco alleging staff at Sutter Health and MemorialCare used Abridge AI during medical visits in past six months to transcribe and transmit confidential physician‑patient communications without clear notice - Ars Technica

wait til you find what happened when doctors dictated notes the last couple decades

Health insurers' AI-driven coverage denials spark class-action lawsuits over biased data - KFF Health News

Courts demand LLM chat logs as federal rulings split on AI medical privilege - Mashable

Almira Osmanovic Thunström created a fabricated eye condition, "bixonimania," and posted two fake preprints in spring 2024 that were ingested by major LLMs, which presented it as a real diagnosis and gave medical advice. - Nature

can I get a pre-approval for treatment tho

AI drug discovery is limited and slow to date - TNW | Opinion

Psychotherapist Lauran Ware spent a month using ChatGPT, finding it offered practical crisis tips but felt emotionally hollow, as Mental Health UK reports one in three UK adults use chatbots and a Brown study warns they mishandle crises. - The Independent

Wharton finds students over-relying on AI scored 17% worse on final tests - Today's Parent

SUNY Binghamton researchers built a four-legged talking robotic guide dog, published at AAAI 2026, that in a seven-participant study scored 4.83/5 for usefulness and achieved 94.8% destination identification in GPT-4 simulations. - StudyFinds

good boy

The faith-based AI market is expanding with projects like Just Like Me's AI Jesus (advertised $1.99/min or $49.99 monthly for 45 minutes), Kyoto University's BuddhaBot, and Magisterium AI trained on 2,000 years of Catholic teaching, amid ethical concerns. - ABC News

giving money to someone pretending to be God is my cultural heritage

Somebody vibe-coded hormuzrun.com - Hormuz Run

Ahead of their time - YouTube

shear madness, sheep in, fluff out AI video - Instagram

Follow the latest AI headlines via druce.ai on Bluesky