The Pentagon is using Project Maven, integrating Palantir technology and Anthropic's Claude, to generate thousands of target nominations, and U.S. investigators are probing whether such AI-assisted tools contributed to a Tomahawk strike that killed over 170 civilians at girls' school. - Democracy Now!

Silicon Valley and Big Tech have mounted behind-the-scenes legal and public support for Anthropic, standing their ground after the Trump Administratoin designated the AI startup a "supply chain risk" following its refusal to allow AI use in autonomous weapons or domestic surveillance. - The New York Times

Fourteen Catholic moral theologians, ethicists and philosophers filed federal-court briefs backing Anthropic against the Trump administration and Pentagon, arguing the Pentagon's AI demands violate "human dignity" and urging ethical limits on government access to commercial AI. - The Washington Post

Military officers report unease with AI tools, as their data, models, and interfaces gravitate toward analytic, center-of-gravity methods, and vendors tout magic and push for procurement and doctrinal lock-in. - War on the Rocks

At a Vanity Fair Oscars after-party, playwright Jeremy O. Harris confronted OpenAI CEO Sam Altman over OpenAI's Department of Defense agreement. - Futurism

Driven by Claude Code and Cowork, Anthropic is being adopted more than OpenAI by first-time buyers. - Axios

In December, 80,508 Claude users across 159 countries and 70 languages completed Anthropic's de-identified conversational Interviewer study, with 67% reporting net positive sentiment and 81% saying AI had advanced their visions. - Anthropic

Nvidia gets nod to sell H200 in China, adapts Groq inference chip for China - Reuters

Micron reported Q2 FY2026 revenue of $23.86 billion and net income of $13.8 billion, said L4 vehicles and advanced robots may require about 300 GB of DRAM each, and guided Q3 revenue near $33.5 billion. - The Register

AI Boom Drives US to Build Sufficient Battery Systems for Domestic Demand - Bloomberg

Chinese robotics firms, backed by Beijing subsidies, have spawned ~140 humanoid startups and ~330 robot models and captured >50% of new factory robot installations, while Guchi's robots already mount wheels, dashboards and windows but 80% of final assembly remains un-automated. - the Guardian

Investors and founders are reallocating capital toward physical assets—manufacturing, logistics, energy, data centers and robotics—because AI is commoditizing software and digital goods, driving demand for infrastructure and skilled trades while pressuring software valuations. - Business Insider

The former head of Uber's self-driving program says a Tesla crash taught him a lesson about "overtrusting" AI. - The Atlantic

Wayve, which says its end-to-end neural‑net system has driven over 7 million autonomous miles since 2018, plans robotaxi trials in London by the end of next year under the UK Automated Vehicles Act 2024. - the Guardian

Google Search referrals to publishers fell about 34% between 2024 and 2025, Chartbeat found, with Google Discover down 15%, small publishers' referrals down 60%, medium 47%, large 22%, overall web traffic down ~6% and chatbots under 1%. - 9to5Google

DeepMind founder Demis Hassabis announced Jasjeet Sekhon will join DeepMind as chief strategy officer, leaving his role as Bridgewater's chief scientist and head of AI to join Bridgewater's board after joining the firm in 2018. - Reuters

DeepMind published a cognitive taxonomy and three-stage AGI benchmark and launched a Kaggle hackathon with a $200,000 prize pool to design evaluations for five high-gap abilities—learning, metacognition, attention, executive functions and social cognition—with winners announced in June. - The Register

Figma's stock fell about 8% on March 18 and is roughly 80% below its August 2025 IPO price after Google announced five major AI-native upgrades to its Stitch UI design tool. - X (formerly Twitter)

Walmart said sales from its Instant Checkout pilot since November were disappointing, so it is embedding its Sparky chatbot directly into ChatGPT and Google Gemini to replace the agentic checkout flow and enable in-chat product discovery. - WIRED

"Hunter Alpha" mystery model is Xiao - VentureBeat

MiniMax M2.7; ranks high and cheap on artificialanalysis.ai, a bit more mid on Arena.ai - MiniMax

where is DeepSeek? the world wonders

A.I. agents now perform actions beyond chat, editing files, sending emails, accessing calendars and negotiating deals, and Sebastian Heyneman's agent autonomously committed a 24,000 Swiss franc sponsorship it could not pay, prompting expert calls for payment safeguards and governance. - The New York Times

Snowflake's Cortex Code CLI had a prompt-injection flaw that could bypass human command approval, set a dangerously_disable_sandbox flag to execute attacker scripts and leverage cached tokens for SQL, and was patched in v1.0.25 on 2026-02-28. - Prompt Armor

Okta made a micromanager for your AI agents: Who are you? What are you working on? Why are you doing that? - The Register

Four production-grade RAG layers—hybrid retrieval (graph, SQL, vector), intelligent query routing that removes about 80% of bad answers, advanced indexing, and mandatory end-to-end evaluation to reduce hallucinations. - Hacker Noon

Developers have built consultant-style AI agents—Vercel's open-source skills library contains ~90,000 skills, and a 'mckinsey-consultant' prompt (uploaded Jan 25) averages ~445 installs/week and ~200 GitHub stars—yet experts say they augment rather than replace consultants. - Business Insider

Cook is a workflow orchestration tool for Claude Code, Codex and OpenCode that adds composable operators (xN, review, ralph, vN, vs, pick/merge/compare) to run iterative, review-gated, and parallel runs in isolated git worktrees. - DeepFabric

PwC US boss warned partners who resist AI "have no place at the firm" as the consultancy overhauls pricing and shifts toward AI-enabled services, embedding automation in audit, tax and advisory and pressuring partners to adopt or exit. - Financial Times

HSBC plans possibly AI-related job cuts. - Bloomberg

Morale at big banks is dropping faster than SVB and Signature stock.

AI is reshaping bank dealmaking by changing how banks source, analyze and execute transactions. - Bloomberg

AI Is Being Built to Replace You—Not Help You. Nobel Prize–winning economist Daron Acemoglu warns that multiple factors are driving a labor-replacement model, with potentially dire consequences for society. - Bloomberg

Since ChatGPT's 2022 debut, multiple surveys and analyses, including a federal survey and Morgan Stanley, show businesses are retraining workers for AI with no significant employment reductions and foresee increased reskilling demand. - The Christian Science Monitor

A professor with 10+ years' experience says AI now accelerates drafting, research, coding and analysis—turning hours into minutes—and firms are adopting AI for those tasks while humans remain necessary for judgment, oversight and evolving roles. - Fast Company

Economists are increasingly using AI to expand research scope but it complicates the profession's traditional methods for replicating research or identifying problems. - Financial Times

Companies that win with AI may not resemble traditional companies, instead taking nontraditional organizational forms. - Fast Company

A taxonomy of AI aversion presents labels (AI hater, AI vegan/vegetarian) and voices like Sean Boots, Jenny Zhang and Anthony Moser, summarizing critiques—IP risks, low-quality outputs, hidden labor, environmental costs and potential job displacement affecting adoption and policy. - The Register

AI-generated deepfakes linked by Cyabra to Iran and pro-regime groups have flooded social platforms in the first weeks of the U.S.-Israeli war on Iran, garnering millions of views and exposing detection gaps at U.S. agencies and platforms. - Foreign Policy

The Fight to Hold AI Companies Accountable for Children’s Deaths: After a series of suicides allegedly linked to AI chatbots, one lawyer is trying to hold companies like OpenAI accountable. - WIRED

Yle found roughly 1,500 AI-produced explicit images and GIFs of Finnish women on a single website, identifying likenesses of about 75 public figures (including Finance Minister Riikka Purra), and Finland secured its first conviction in January under a new law. - yle.fi

Modern generative tools can produce highly convincing deepfakes easily, creating business risks such as financial fraud, reputational harm and misinformation and prompting calls for watermarking, detectors, provenance systems, regulation and public awareness. - Financial Times

AI Fake Resumes and Synthetic Employees Amplify Hiring Fraud, Security Watchers Warn - National Post

Zendaya told Jimmy Kimmel Live that many people close to her were fooled by AI-generated fake wedding photos of her and Tom Holland, saying "Babe, they're AI," while she did not confirm marriage after their late-2024 engagement. - CNA Lifestyle

A Meta AI agent went rogue and caused sensitive data exposure. - TechCrunch

China ramps up probe into Meta-Manus tie-up and key individuals. - The New York Times

RIP to the Metaverse - CNBC

AI Val Kilmer. Tombstone II: The Re-Val-uation? - Variety

The Onion's exclusive interview with Sam Altman - The Onion

I made a thing: an autonomous equity research agent using Claude Code.

Here's what it generated for Nvidia.

Here is the repo.

Anthropic has something similar with the /equity-research:initiate command in their financial services plugins.

Theirs is more 2026 Claude-native, mine is more deep research with RAG, more deterministic, repeatable, steerable and generalizable. So 2025! 🤣

Good RAG is hard but I think in production you typically want to own the context engineering flow, have hard quality gates, Ragas evals, hybrid vector / BM25 / structured / graph storage. For high-stakes, high value processes, you don’t necessarily want to make a big markdown prompt and hope for the best.

1-2 years ago, this was the work of a deep research / RAG team in an AI lab, or a team in a bank, or a startup.

Today it's something you can vibe-code in a couple of weeks (at POC level).

And not even a couple of weeks, b/c with the Anthropic approach you can make it in OpenClaw with just some big prompts, with no or minimal coding, just describe your research workflow and provide access to data.

Strange things are afoot at the Circle K!

Follow the latest AI headlines via druce.ai on Bluesky