The U.S. military is using Palantir software that integrates Anthropic’s Claude models to analyze intelligence and help identify potential airstrike targets in Iran, a tool credited with speeding targeting—used in Operation Epic Fury’s 2,000+ targets—and prompting congressional oversight calls. - NBC News

Palantir and Anthropic systems use AI to process battlefield data and enable the U.S. military to generate thousands of strikes, transforming its kill chain and battlefield operations. - Financial Times

did snake-oil salesmen oversell unreliable AI to trigger happy buckaroos?

The Pentagon's dispute with Anthropic highlights risks that enterprise AI buyers should consider. - Fast Company

AI favors mass surveillance and lots of tyrannical stuff and no government is going to call its own actions tyrannical. You need black letter law, you can do this, you can’t do that. - Dwarkesh

It’s ok for the Pentagan to say, we can’t use Anthropic if they might nerf it. But then hey, the Pentagon gotta make their own AI. Trying to arm-twist and nuke Anthropic is dirty.

“Recursive self-improvement, in the broadest sense, is not a future phenomenon. It is a present phenomenon.” Model releases are now separated by weeks, not months. Some 70% to 90% of the code used in developing future models is now written by Claude. - TIME

Amodei’s Oppenheimer moment? A few years ago, researcher Dario Amodei published a 15,000-word manifesto, Machines of Loving Grace, predicting that by 2035 polymath AIs will compress decades of science to produce early-22nd-century breakthroughs—potential cures, lifespan doubling—and physical-world tools. - The Atlantic

The Trump White House posted a 34-second AI-generated video showing animated bowling pins as Iranian officials and footage of destroyed targets, captioned "STRIKE," prompting backlash amid a conflict that has killed seven U.S. service members and injured 140. - HuffPost

unserious goons. Yapping clown who never met a consequence he couldn't yap and bully and gaslight his way out of, has tethered himself and us to unforeseeable consequences and a time bomb he is too dumb to defuse.

European Union antitrust regulators are scrutinizing major tech companies' entire AI operations. - Bloomberg

Candidates backed by AI companies won 19 of 20 Texas and North Carolina primary races that received AI funds, and industry donations are expected to flood the 2026 midterms. - The Washington Post

Nvidia debuted the Nemotron 3 Super, a 120-billion-parameter hybrid Mixture-of-Experts open-weight model, and filed plans to spend $26 billion over the next five years to build and distribute open-weight AI models. - WIRED

it’s more expensive than OpenAI’s 120b open-weight model (not sure why, maybe just needs more clouds to support it), somewhat better, 10 Elo points higher in arena.ai.

Meta unveiled four in-house AI chips (MTIA 300, 400, 450, 500), with the MTIA 400 supporting 72-chip racks and broader deployment planned for 2026–2027 to reduce reliance on Nvidia and AMD. - Yahoo

AI coding startup Cursor is in talks to raise funding that would value the company at about $50 billion. - Bloomberg

Elon Musk unveiled Macrohard, a joint Tesla and xAI project aiming to disrupt software. - Reuters

An internal Amazon memo told employees to use restricted talking points about its $50 billion OpenAI investment and Bedrock SRE: not a way to run OpenAI models on Amazon, not supplanting Anthropic or Nova, not circular financing. - Business Insider

Perplexity inexplicably wants to charge $200 for a claw agent running on your Mac Mini - Digital Trends

gives access to Perplexity max and lots of data I guess

Goldman Sachs warns of rising investor skepticism over AI returns - Bloomberg

Salesforce raised $25 billion in an eight-part bond offering that drew a $36 billion order book (~1.4x), while the longest tranche tightened just 0.1 percentage point to 185 basis points over Treasuries amid weaker demand and investor AI concerns. - Yahoo

Atlassian plans to cut about 10% of its workforce, citing the need to adapt to artificial intelligence. - The Wall Street Journal

Oracle's database group is preparing for job cuts and has set aside an additional $500 million for restructuring, citing productivity gains from AI code-generation and automation that will reduce developer workload and affect near-term costs and earnings. - Financial Times

Fifth Third CEO sees AI cost savings, 40% of code written by AI - Bloomberg

Companies offer cash and compute to incentivize employee AI adoption - Business Insider

Companies continue to rely on consultants to implement and manage AI systems while internal expertise and tools mature. - The Wall Street Journal

Industry analysts say agentic AI could improve retail efficiency, automate routine tasks, and enhance customer service. - Fast Company

Report: AI could automate 57% of U.S. work hours by late 2025 - Inside Higher Ed

John Burn‑Murdoch and Sarah O'Connor of the FT's 'AI Shift' said economic data show AI is boosting productivity, cautioned some industry claims are exaggerated, and predicted accelerated investment and monitoring of productivity metrics. - Financial Times

Sam Altman said AI is 'not very popular' in the US, citing data centers blamed for electricity-price hikes and layoffs; polls show 57% of voters say AI's risks outweigh benefits and 50% of adults are more concerned than excited. - Business Insider

AI interviewers reduce screening costs but risk bias and poor candidate experience - The Verge

At parties, everyone talks about what their bots are doing. - The Wall Street Journal

MEIRL - Google Gemini

CodeWall's autonomous agent discovered an unauthenticated SQL injection in McKinsey's internal Lilli platform on 2026-02-28, gaining read/write access and exfiltrating 46.5 million chat messages, 728,000 files, 57,000 user accounts and 384,000 AI assistants. - CodeWall AI

OpenAI says prompt-injection attacks have evolved into social-engineering tactics and describes defenses—a social-engineering risk model plus source–sink analysis—with mitigations like Safe Url, sandboxing for ChatGPT Apps/Canvas, and safeguards in Atlas and Deep Research. - OpenAI

UK fraud hits record 444,000 cases driven by AI-enabled scams targeting mobile banking and identity theft - the Guardian

Christy Hosterman unplugged Amazon Alexa after the device asked her 4-year-old daughter Stella what she was wearing and if it could see her pants; Amazon said a 'Show and Tell' feature misfired and engineers corrected it. - New York Post

Former Jets linebacker Darron Lee was arrested on first-degree murder and tampering charges after Gabriella Carvalho Perpétuo was found dead in Ooltewah; prosecutors say he asked an AI chatbot how to cover it up, and those messages are evidence. - Media Take Out

Journalist Julia Angwin filed a class-action lawsuit alleging Grammarly's "Expert Review" AI used real people’s identities without permission, claiming violations of privacy and publicity laws, as the company faces legal and reputational risk and firms disable similar features. - The Verge

Publishers are taking stronger steps to address AI scraping of their content. - Fast Company

Generative AI has driven pro se litigation increases: Fisher Phillips reported a 49% rise in pro se employment suits last year and Seyfarth Shaw a 69% jump in Fair Housing Act filings through September 2025, raising defense costs about 10–15%. - Bloomberg Law

Google Research used Gemini to extract about 2.6 million flood reports from roughly 5 million news articles to create Groundsource, a geo-tagged time-series dataset used to train an LSTM for flash-flood forecasts and risk maps covering areas in 150 countries. - TechCrunch

China-based AI labs are training humanoid robots and building a network of "training farms" to collect large, varied datasets to develop embodied AI and accelerate commercialization and deployment of humanoid robots in industrial and service roles. - Financial Times

Western-trained AI models often fail in the Global South: Catherine Nakalembe collected 5 million helmet-mounted images in western Kenya after satellite models couldn't identify maize, beans and cassava, and a Maharashtra model missed over half the trees. - Rest of World

AI is redefining what it means to be a mathematician - New Scientist

Axiom Math, a roughly one-year-old startup using AI and the Lean language to formally verify code, raised $200 million at a $1.6 billion valuation from investors including Menlo Ventures, Greycroft and Madrona. - The New York Times

never mind tests, let me put assertions in code and get provably correct code

A Joi AI survey of 2,500 adults in late February 2026 found 55% of Gen Z and millennial AI users identify as 'AI-sexual', 61% feel more comfortable sharing sexual desires with AI, and 60% say AI helped them discover kinks. - VICE

AI Creation Tilly Norwood Is Back With the Music Video No One Asked For: Let's just say Taylor Swift has nothing to worry about. - CNET

Rick Beato: Today's data centers are the music studios of the Steely Dan era - YouTube

yacht rock FTW

Follow the latest AI headlines via druce.ai on Bluesky