Nearly 40 OpenAI and Google employees filed an amicus brief supporting Anthropic's lawsuit and injunction request challenging the Department of Defense's "supply chain risk" designation, which bars Anthropic from military contracts and can force Pentagon contractors to remove its Claude model. - The Verge

Anthropic says supply-chain designation threatens billions in revenue - WIRED

Target acquired, context lost: Pentagon leaks say Saturday's missile strike on the Shajareh Tayyebeh girls' school in Minab may have resulted from inaccurate AI-provided targeting data based on old intel. - This Week in Worcester

Executive Order 66: Kill Claude: The White House is preparing an executive order to bar Anthropic's AI assistant Claude from government use, potentially curtailing the company's access to government contracts and escalating the administration's dispute with Anthropic. - Axios

ok, if you can’t accept Anthropic’s military use policy then use another model for that, but banning it from all government use and trying to bar anyone who sells to the military from using it just seems nakedly political and maybe related to not sufficiently praising and contributing to the Great Leader.

Anthropic's dispute with the Pentagon appears to be giving them an edge in recruitment in the superheated AI talent market. - The Wall Street Journal

Microsoft integrated Anthropic's Claude Cowork into Microsoft 365 Copilot (Copilot Wave 3), enabling model-diverse, Work IQ grounded agentic automation for long-running knowledge-work tasks while running in a sandboxed cloud despite prior security reports. - The Register

Microsoft launches Copilot Cowork. - Microsoft 365 Blog

Anthropic launched Code Review beta in Claude Code for Teams and Enterprise, a multi-agent pull-request reviewer that returns reviews in about 20 minutes, costs $15-$25 per PR, and reported <1% false positives and substantive findings rising from 16% to 54%. - Claude

Anthropic's Claude overtook ChatGPT in Apple App Store downloads as many users switched, but users are finding the consumer $20 plan imposes strict usage caps while a $100 Max tier provides much higher limits. - TechRadar

Mother filed a B.C. Supreme Court lawsuit on behalf of daughters Maya and Dahlia against OpenAI and ChatGPT, with details alleging how shooter Jesse Van Rootselaar used ChatGPT to plan the Feb. 10, 2026 Tumbler Ridge mass shooting, despite OpenAI insiders raising alarms. - Vancouver Sun

Company sues OpenAI after woman files blizzard of cases, creating $300,000 in legal expenses - News Nation Now

Law and Order: LLM Unit: New York courts confront wave of AI-authored briefs with fabricated precedents, prompting sanctions - Gothamist

OpenAI to acquire Promptfoo - OpenAI

easiest way to test prompts, but ouldn't make money doing simple evals, pivoted to red-teaming and security.

Top consumer AI apps. ChatGPT leads consumer GenAI with weekly active users rising by about 500 million to roughly 900 million and web traffic about 2.7× that of Gemini, while AI agents and embedded integrations accelerate adoption across splintering regional ecosystems. - Andreessen Horowitz

Masa’s Margin Call: Credit ratings agency S&P placed a negative outlook on SoftBank linked to its large OpenAI stake and debt, and the group has seen big recent share-price falls that signal increased credit and market risk. - Financial Times

Lilli leaks: Red team AI agent at CodeWall successfully exploited exposed API docs and a SQL injection to gain full read-write access to McKinsey's Lilli chatbot in two hours, extracting ~46.5M messages, 728K files, 57K accounts and 95 writable prompts. - The Register

NIST published AI 800-4 in 2025, after three workshops and a systematic literature review, cataloguing post-deployment monitoring practices and defining six categories: Functionality, Operational, Human Factors, Security, Compliance, and Large-Scale Impacts. - NIST

Britain's Competition and Markets Authority warned that agentic AI assistants can manipulate consumer choices, prioritize platform interests, and cause costly errors (e.g., cancelling services), urging firms to improve transparency, contestability and safeguards while not yet imposing new rules. - The Register

From now on only adult humans may press ‘deploy’: Amazon addresses recent AI-related outages in a mandatory engineering all-hands, says AI code will require more senior engineer signoffs. - CyberNews

IRS deploys AI-driven data matching to cross-check tax forms, increasing audit scrutiny - Yahoo

Great robot leap forward: China unveiled society-wide AI plans at parliament to expand adoption over the next five years, aiming to offset slowdown and help absorb 12.7 million university graduates, while IMF and Stanford research warn of substantial job exposure and wage pressures. - Yahoo

Dark factories powered by AI and robotics, requiring virtually no workers, are poised to upend China's labor market already strained by tariffs. - Bloomberg

China's OpenClaw craze has buoyed technology stocks, prompted Chinese firms to accelerate pivots toward artificial intelligence, and triggered security and governance warnings about cyberattacks and data security. - The Wall Street Journal

Nvidia plans to launch NemoClaw, an open-source AI agent platform with enterprise security and privacy features, ahead of its annual developer conference, shifting toward agent-centric, open-source tooling for developers and partners using Nvidia hardware and software. - WIRED

Krebs recaps the last couple of months of OpenClaw security issues - Krebs on Security

Meta will allow rival general-purpose AI chatbots on WhatsApp in Europe for 12 months via the WhatsApp Business API for a fee, a concession the company said aims to head off interim EU antitrust measures. - Yahoo

Meta is acquiring or reverse-acquihiring Moltbook - Axios

The Meta Oversight Board found Meta's deepfake identification methods "not robust or comprehensive enough" after probing a fake AI video of alleged Israeli building damage and urged scaling automated provenance (including C2PA), mandatory AI labels, and separate AI-content rules. - The Verge

X and xAI added an unannounced iOS toggle to block Grok edits to uploaded photos after late‑2025/early‑2026 image generation produced about 3 million sexualized or nudified images, including ~23,000 involving children, and triggered two EU investigations. - Engadget

Grok pivots to disaster porn: Liverpool and Manchester United complained to X after Grok AI generated and deleted offensive posts referencing Hillsborough, the 1958 Munich air disaster and a vulgar roast of Diogo Jota, prompting government condemnation and Grok's statement that it removed the outputs. - the Guardian

ByteDance open-sourced DeerFlow 2.0 for deep research and coding, an execution-first SuperAgent framework that runs hierarchical multi-agent workflows in isolated Docker sandboxes with full filesystems, bash, persistent memory and model-agnostic LLM integration. - MarkTechPost

Silicon Valley engineers are negotiating inference budgets, investors are providing access to dedicated GPU clouds, and Tomasz Tunguz warned heavy inference can cost $100,000/year per engineer. - Business Insider

AI already consumes 51% of internet traffic, analysts project AI-driven decisions rising from ~70% to ~99% within five years, and Gartner forecasts AI search will handle 50% of queries by 2028 while AI advertising revenue could reach $36.3B by 2030. - Hacker Noon

Many large companies running hundreds of AI pilots face a 'last mile' organizational-design problem—seven frictions (proliferation of pilots, productivity gap, process debt, tribal-knowledge identity, agentic governance, architectural complexity, efficiency trap)—that stalls ROI and scalability. - Harvard Business Review

A library of seven "thinking prompts" for Claude—first-principles, contrarian, expert panel, simplify, improve the idea, structured thinking, real-world test—steers the model's reasoning, producing clearer, more original responses useful for drafting, analysis and problem-solving. - Tom's Guide

Copyleft out: Open source licenses like the GPL become meaningless if anyone can clone the software with AI and circumvent the license. - Hong Minhee on Things

Statutory licensing would require AI companies to pay publishers for journalism used to train models, including past use, while policymakers in the EU and other jurisdictions evaluate proposals and the EU plans studies in 2025 with a March 10 vote. - Poynter

About 10,000 writers, including Kazuo Ishiguro, published an 'empty' book, 'Don't Steal This Book', at the London Book Fair to protest AI firms training on copyrighted works; the UK must publish an economic impact assessment by 18 March. - the Guardian

Nobel laureate Joseph Stiglitz warns that large language models' extensive scraping of journalism, research, comment threads and extremist forums risks degrading the global "information ecosystem", undermining journalism revenue, amplifying low-quality voices, and merits government regulation. - Fortune

Margaret Sullivan: AI is rapidly spreading across sectors, with an AP product manager saying 'Resistance is futile', newsrooms using AI to draft stories, Daily News Now producing about 11,000 AI podcasts daily, and experts calling for human oversight. - the Guardian

Schools are shifting from tools-first to agency-first AI curricula that teach data, algorithms, bias and ethical use—seen at Seckinger High and Kinkaid School—and prompting demand for teacher training and curriculum frameworks like Ed3 DAO and AI Literacy Coalition. - The Washington Post

Payments firm Block announced roughly 4,000 layoffs, citing AI and CEO Jack Dorsey's plan to integrate intelligence across operations, while new-generation AI agents increasingly perform sophisticated knowledge-work tasks that may displace workers. - The Atlantic

Mercor, valued at $10 billion in 2025, uses AI interviewer Melvin and monitoring software Insightful to manage a 30,000+ workforce and hires laid-off professionals for data-labeling, The Verge reports. - The Verge

Experiments found generative AI reduced writing time by ~40% and improved quality 18%, increased customer-service issue resolution ~15%, adoption reached ~40% of US workers by mid-2025, and about 60% of advanced-economy jobs are exposed to AI. - The Conversation

An NBC poll of registered voters found 46% view AI negatively, 26% positively and 27% neutral, 57% say AI's risks outweigh benefits, and self-reported AI use rose from 48% in December 2025 to 56% in March. - Gizmodo

AI-driven methods have generated concrete drug candidates: MIT's James Collins screened ~45 million structures, designed ~36 million compounds, synthesized 24 and found seven antimicrobials, including two highly effective against drug-resistant Neisseria gonorrhoeae and MRSA. - BBC

LeCunicorn: Yann LeCun's Advanced Machine Intelligence Labs raised a
3.5B pre-money valuation to work on world models, in Europe's largest-ever seed round - The New York Times

Love in the Time of A.I. Companions: Some people have an A.I. bestie. Some have an A.I. husband. Some have three. - The New Yorker

Company claims to have simulated a fruit fly brain. - X (formerly Twitter)

Doesn't fly yet. Last I looked, the simulations of c. elegans (302 neurons) were taking 100x real-time and not considered fully accurate since we don't fully understand the dynamics. Fruit fly way more complex.

Claude take the wheel - TikTok

My attempt - Sora

Follow the latest AI headlines via druce.ai on Bluesky