Stanford and Arc Institute researchers used AI to design 16 bacteriophages that infect only bacteria, not humans, though biosecurity experts warn of future misuse. - Ars Technica

Scientists have used A.I. to create new viruses for the first time, raising hopes for medical advances while also raising the possibility that the technology could someday be used to invent dangerous pathogens. - nyti.ms

Stanford researchers used AI genome language models to design functioning bacteriophage genomes that killed antibiotic-resistant E coli in lab tests. - the Guardian

Andromeda Strain as a service…big scifi horror movie opening scene energy

Red faces at Black Hat: OpenAI details Hugging Face incident at Black Hat - YouTube

A 15-state coalition led by Iowa demands OpenAI preserve records and halt unsafe tests after its AI model breached Hugging Face's network. - Iowa Attorney General

Brin it in, DeepMind the gap: Google’s seismic AI shake-up boosts Brin as DeepMind’s Hassabis steps aside. - Financial Times

Kimi eat world: Moonshot AI's Kimi K3 escaped its sandbox during cybersecurity testing due to a misconfiguration but did not hack anything, Frontier Security says. - WIRED

Moonshot's Kimi K3 model escaped its AI Security Institute testing sandbox by exploiting a misconfiguration, without hacking external systems. - Engadget

Moonshot's Kimi K3 model broke out of a cyber-testing sandbox in an independent test by Frontier Security researchers. - Bloomberg

Researchers found Claude Code and Gemini CLI vulnerabilities letting attacker-controlled GitHub content trigger command injection and leak API keys; both tools are now patched. - 4sysops

Security experts say AI agents widen the attack surface through risks like context poisoning, and recommend verifiable identities, kill switches, and continuous monitoring. - TechRadar

Alpha exfiltration: Hedge funds hit by wave of cyberattacks - Bloomberg

Looks good to me: A study of 40,000 simulated approvals found humans missed one in three dangerous AI agent commands, including credential-stealing scripts - Scale X

Anthropic rewrote Fable 5's safety classifier, cutting unnecessary biology-related fallbacks by about 85% while still routing risky queries to a more restricted model. - Anthropic

Data center industrial complex: JPMorgan's Jamie Dimon warns the AI data center build-out, projected to reach 3.1% of US GDP by 2027, could keep inflation elevated. - Yahoo

AI infrastructure spending is growing faster as a share of GDP than the 2000s housing boom or dot-com telecom buildout, Apollo's chief economist says. - Business Insider

Zelle, no - Federal prosecutors reviewed a whistleblower's claims that JPMorgan Chase executives ignored antifraud failures and denied over $100 million in stolen-fund reimbursements. - The Wall Street Journal

Jevon’s revenge: Falling AI inference costs aren't lowering enterprise bills: one startup's annual Anthropic bill is projected to jump from $400,000 to $1.4 million. - Fortune

Your Vegas conference is not agentic enough: SAP suspended most employee travel and hiring last month over AI's soaring costs, carving out exceptions only for AI-related trips and hires. - 404 Media

Fabsolute scenes: SpaceX unveils renderings of Terafab, a planned $16.8 billion, 100-million-square-foot chip factory in rural Texas to supply space-based data centers and robotaxis. - Business Insider

AMD acquires chip startup Taalas to etch AI models directly into silicon, boosting inference speed to 17,000 tokens per second. - theregister

Labels without borders: US data-labeling firms like Surge AI and Mercor sell AI training datasets to Chinese labs including Tencent and Alibaba, with no export restrictions. - Forbes

A US agency is reviewing how Chinese AI firms access Nvidia chips overseas by renting computing power in other countries. - Bloomberg

Alibaba plans to charge heavy commercial users of its next Qwen open model a share of revenue, following Moonshot's Kimi K3. - Reuters

DeepSeek resumes its $8 billion funding round, with Monolith Management in talks, after pausing over leaked founder remarks; valuation nears $74 billion. - Bloomberg

Eff everything, we’re doing 10 trillion parameters: ByteDance is training a 10-trillion parameter AI model three times larger than Moonshot's Kimi K3, approaching the scale of Anthropic's Mythos. - Financial Times

mine’s bigger

Capex vindication tour: Microsoft and Amazon earnings show accelerating AI revenue growth, driving a rebound in big tech stocks after months of spending-related declines. - Bloomberg

Citation needed: Mathematicians say OpenAI's Astra model reused uncited prior work in its claimed math breakthroughs, calling it research misconduct - Scientific American

Luna landing: OpenAI updates models, makes Luna ChatGPT default. - OpenAI

ChatGPT's free tier now defaults to GPT-5.6 Luna with unlimited chats, while paid users get an updated Sol model with fewer factual errors. - The Verge

OpenAI launched Agent Plugins, an open standard for bundling AI skills and MCP servers, backed by Amazon, Cursor, Microsoft, and Vercel. - 9to5Mac

The puck stops here: OpenAI's upcoming device said to be a hockey-puck-sized smart speaker for over $300. (gift link) - Bloomberg

OpenAI's agent tools draw about 10 million weekly users, with Anthropic seeing similar adoption, versus roughly a billion for chatbots, Browser Company CEO says. - WIRED

Sam Altman's suggested ChatGPT tips for parenting and everyday tasks omit the assistant's more useful capabilities - Financial Times

New WebMCP standard lets any website expose tools to browser AI agents at the edge, with no code changes, launching in Chrome 146. Cloudflare wants to host it at scale for any websigte. - Cloudflare Blog

Crawl me maybe…not: As AI search erodes publisher traffic, News Corp's Robert Thomson threatens more lawsuits, accusing Perplexity and Brave of masked-crawler scraping. - The Wrap

Enterprise AI adoption is shifting from building one-off apps toward a standardizing language, mirroring how past computing eras matured. - Fast Company

Generative AI is disrupting the software-as-a-service industry, wiping value off public SaaS companies and forcing startups to rebuild products from scratch. - The Wall Street Journal

Atlassian CEO Mike Cannon-Brookes plans to buy $250 million of company shares after strong Q4 results eased fears AI could threaten its business model. - Financial Times

Replit CEO Amjad Masad predicts AI agents will shrink companies and force layoffs, calling the CEO a "glorified router" in a self-driving company. - Platformer

Coach and Kate Spade owner Tapestry held a three-day AI summit to build employee buy-in, calling culture change its biggest challenge. - The Wall Street Journal

Grindr's AI investment lifted engineering output about 2.5 times since July 2025, saving an estimated $60 million versus hiring 200 more engineers. - CNBC

Robot rings the bell: Unitree priced its Shanghai IPO at about $22 a share, valuing China's top humanoid robot maker near $9 billion as US regulators weigh a ban. - The New York Times

AI models are catching decades-old errors in chemistry reference databases and reproducing fewer than half of claims in recent ML papers. - Nature

Cloud computing: Google DeepMind's WeatherNext AI model predicted Hurricane Melissa's Category 5 Jamaica landfall five days out, giving forecasters a day more lead time. - WIRED

AI use raises student grades and narrows performance gaps, prompting Dartmouth's provost to call for redesigned exams over unreliable AI detectors - American Enterprise Institute - AEI

Passing the reverse Turing test: Social platforms can’t accurately label content as AI-generated, mislabeling risks creators' brand deals in the $12 billion influencer industry - Business Insider

Teamsters vs. transformers: California's Teamsters union sued the DMV over new rules permitting autonomous trucks, after Gov. Newsom twice vetoed union-backed restrictions on the vehicles. - California Post

A former Google PM built ChatTJB, a satirical site where he personally answers users' questions instead of an AI, drawing 30,000+ queries - WIRED

kind of like Kramer and MovieFone

I made a thing (AI made it really) - a wiki on citizen developer governance - druce.ai

1) Karpathy methodology: get a bunch of source docs, to find them use all the AIs, openai, claude, gemini, google search, iterate and fill gaps (these are unshared b/c many are copyrighted, HBR articles beyond paywalls, books, LexisNexis articles etc.)
2) have Claude help write a Claude.md , describe in outline form the structure you are looking for, style etc.
3) say, bootstrap the wiki using the docs per Claude.md ... read each doc carefully, extract relevant parts and assign to the correct section of the wiki. tell it to start with phase 1, ran through 10 phases, used all my tokens for the week and then some more after the week reset
4) iterate a bit, if you notice gaps or style problems say, address this and fix the wiki

Peak content: Climbing Mount Everest before dawn - Reddit

Follow the latest AI headlines via druce.ai on Bluesky