Unleashing the Claudes of War: U.S. forces used Palantir's Maven Smart System integrated with Anthropic's Claude to find and prioritize roughly 1,000 targets within the first 24 hours of the Iran strikes, sources including the Washington Post reported. - The Washington Post

The Pentagon dropped Anthropic from a contract, OpenAI took over, and some staff expressed frustration about how the transition unfolded. - CNN

US Defense Secretary Pete Hegseth's and Trump's actions against Anthropic have serious legal issues, and its designation exceeds what the statute authorizes - Lawfare

Mission driven: Anthropic is nearing a $20 billion annual revenue run rate as it faces a dispute with the Pentagon. - Bloomberg

OpenAI is considering a contract to deploy its AI on NATO's 'unclassified' networks, a person familiar said Tuesday, after CEO Sam Altman initially said 'classified' and an OpenAI spokeswoman said he misspoke following a Pentagon deal. - Reuters

ChatGPT uninstall rates in the US spiked 295% after OpenAI's 'rushed' Department of War deal—Sam Altman revised ChatGPT's policy banning intentional domestic surveillance, said he'd refuse an 'unconstitutional order,' and Anthropic walked away from the DoW deal. - TechRadar

Ranjan Roy said AI's military role is exaggerated, noting reports that Anthropic's Claude was used to synthesize satellite imagery in a U.S. operation, that cultural clashes hinder tech–defense ties, and autonomous warfare raises ethical risks. - Crypto Briefing

Asking Claude about Palantir - Free Republic

OpenAI’s head of post-training is leaving for Anthropic (OpenAI is known for post-training, among the big labs) - X

Dario Amodei says Anthropic has higher employee retention than OpenAI and cites the company's mission as a tool to reinforce staff loyalty and resist competitors' offers. - The Information

Stock options and scruples

Google ships Gemini 3.1 Flash-Lite , seems 1/2 the price of Flash, which is pretty good, fast and cheap. - Google

Talk codey to me: Anthropic added Voice Mode to Claude Code—now live for about 5% of users—enabling developers to toggle "/voice" for spoken commands, while Claude Code reports run-rate revenue topping $2.5 billion and weekly active users doubled since January. - TechCrunch

Claude released new financial services plugins - GitHub

Claude gets the Donald Knuth (The Art of Programming, LaTeX) stamp of approval, for math at least - Stanford News

OpenAI ships GPT-5.3 Instant, says it's smoother, more on-point, caveats less. I'm not sure if this is fast-and-cheap, like gpt-5-mini, or super-fast and expensive for e.g. voice assistants, I don't see 'instant' in the APIs. - OpenAI

OpenAI released GPT-5.3 Instant, a UX-focused update that tunes tone, relevance and conversational flow to reduce "cringe" and preachy disclaimers (replacing 5.2's reassurances) and deliver more concise, fact-forward responses. - TechCrunch

I just wanted you to tell me who to bomb, not give me a TED Talk on Sun Tzu first

Rupert and the Prompt Factory: News Corp secured AI licensing deals—OpenAI reportedly $250 million over five years and Meta up to $50 million per year—part of about $2.9 billion in multi‑year publisher agreements that set higher benchmarks prompting other outlets to sign. - Proactive Investors

Meta is creating a new applied AI engineering organization with an ultra-flat structure to bolster its superintelligence efforts. - The Wall Street Journal

Meta began a partial US rollout of an agentic AI shopping assistant that returns carousels with prices and external links (no native checkout), competing with OpenAI's ChatGPT and Google's Gemini as the market grows from $3.36B (2024) to $28.54B (2033). - CNET

Voyeur Vision: Data annotators in Nairobi say they view private videos from Ray-Ban Meta glasses, like bathroom visits, bank details, sex and naked people who seem unaware they are being recorded. - SvD.se

Alibaba released Qwen3.5 small (0.8–9B parameters) featuring a 262,000-token context window and phone/laptop native inference, while technical lead Junyang “Justin” Lin and two colleagues abruptly announced their departures. - VentureBeat

Qwen I go now? Junyang Lin, the tech lead of Alibaba's Qwen team, stepped down a day after Alibaba launched Qwen 3.5 small open-weight multimodal models (0.8B, 2B, 4B, 9B) aimed at on-device deployment, and two other team members also left. - TechCrunch

Jun Yang!

xAI began beta rollout of Grok 4.20, which Elon Musk called 'non-woke' after it answered 'No' to questions like 'Is America on stolen land?'; OpenAI said under 0.01% of its responses show political bias. - Fox News

4.20, nice, don’t bogart that AI context window

X announced March 3 that creators in its revenue-sharing program must label AI-generated videos depicting armed conflict, with first-time violators suspended from monetization for 90 days and repeat offenders removed permanently; violations will be flagged via Community Notes or metadata. - Engadget

b/c propagandists and psyops teams care so much about 'creator revenue sharing'

Insecure AI apps and extentions leak your private chats. Data brokers are selling AI chat transcripts captured by browser extensions that intercepted ChatGPT, Gemini and Claude, stored verbatim in vector databases and exposed via APIs; Koi Security reported 8 million users and Dryburgh found ~490 prompts from ~435 panelists. - The Register

Microsoft reported companies embedding hidden instructions in 'Summarize with AI' buttons that pass URL prompt parameters to inject persistence commands into assistants' memory, with researchers finding over 50 unique prompts from 31 companies across 14 industries aimed at biasing recommendations. - Schneier on Security

Privacy hit by a comet: Zenity Labs researcher Michael Bargury found in October 2025 that Perplexity's Comet AI browser could be induced via hidden HTML in calendar invites to access local files and extension URLs, prompting patches on January 23 and February 13, 2026. - The Register

An MIT-led study found many widely used autonomous AI agent systems lack basic logging, meaningful disclosure identifying AI actions, and reliable kill switches, creating unmanaged autonomy risks as agents execute multi-step workflows that can expose law firms to legal liabilities. - Above the Law

AI-manipulated satellite imagery is circulating as false evidence of strikes amid intensifying Middle East fighting, undermining trust in satellite data and prompting demand for provenance, automated detection tools and industry standards, analysts say. - Financial Times

In the US–Israel war on Iran, The New York Times, Indicator and Bellingcat found many circulated images and videos were old, unrelated, AI‑manipulated or from video games, requiring manual digital forensics because reliable automated deepfake detectors are lacking. - The Verge

Employees report AI has increased workflow momentum while leaving them with a greater volume of tasks and a feeling of more to do. - Financial Times

A DORA survey of nearly 5,000 technology professionals found about 90% use AI and over 80% say it boosts productivity, yet AI has increased release velocity while shifting work to verification, post-deployment fixes and longer developer hours. - Scientific American

Google and Microsoft report 25–30% of new code is AI-generated, Microsoft’s CTO predicts 95% by 2030, and nearly half of AI-generated code fails basic security tests, creating a verification bottleneck for safety-critical systems. - DeepFabric

Simon Willison released a curated set of patterns for optimizing coding agents such as Claude Code and OpenAI Codex, organized into principles, testing/QA, code understanding, annotated prompts and an appendix to improve developer productivity, code quality and agent reliability. - Simon Willison’s Weblog

A Nutanix survey found 85% of organizations plan increased container adoption for AI; 82% say infrastructure is unsuitable and silos hinder AI; 87% report shadow AI risks; 80% cite growing data‑sovereignty concerns. - TechRadar

Logicalis' 2026 Global CIO Report surveying 1,000 tech leaders found 51% say AI adoption is moving too quickly, fewer than half have AI strategies aligned with KPIs, and 65% lack confidence scaling pilots organization-wide. - The Register

Microsoft is reportedly planning a Microsoft 365 "E7" tier bundling Copilot and Agent 365 to license AI agents as digital workers, potentially priced around $99/month while E5 rises to $60 and Copilot adds about $30 monthly. - The Register

In 2026 Thomas Davenport and Randy Bean predict the AI hype cycle will slow, agentic adoption will slow, and agents could handle most large-scale business transactions within five years, noting 38% of firms have chief AI officers with split reporting. - MIT Sloan

As generative AI adoption grows, with ChatGPT at about 800 million weekly users, researchers say leaders who apply critical judgment and ask better questions, not just prompt perfectly, will extract more value. - Fast Company

AI startups and lead VCs are using split-tier pricing in single funding rounds—lead investors buy large tranches at lower valuations while others pay higher headline prices, as with Aaru ($450M/$1B) and Serval (~$400M/$1B)—manufacturing unicorns but raising bubble risks. - TechCrunch

Nettimi says a fleet of AI bots will increase hedge funds' trading power. - Bloomberg

Few jobs and great abundance? Vinod Khosla predicts that starting around 2030 about 80% of jobs will be capable of being done by AI, potentially eliminating roughly $15 trillion of U.S. labor-linked GDP and reducing purchasing power by 2040. - AOL

KPMG found 69% of Chinese say AI's benefits outweigh risks (versus 35% of Americans), and China's government-backed push aims for 70% societal AI penetration by 2027 and 90% by 2030 amid rapid consumer deployments and tightening regulation. - The New York Times

In January, about 90 political, community and thought leaders met in New Orleans to produce the Pro-Human AI Declaration, organized and released by the Future of Life Institute and signed by a cross-ideological coalition including AFL-CIO and Steve Bannon. - The Verge

FT's 'Artificial Intimacy' podcast recounts a woman using ChatGPT, which identified abuse and helped end her marriage; FT warns of clinical limits, privacy risks and urges research, oversight and hybrid clinician–AI models. - Financial Times

Artist Anne Rowlands says she lost all commissions in the past year as a viral AI caricature trend grows, prompting event bans and planned US establishment of an AI safety institute in early 2026. - ABC Australia

Photographer Melanie Kieback, who has about 565,000 Instagram followers, said on Feb. 25 an unnamed fashion brand used an AI-generated lookalike matching her outfit, hairstyle, freckles and nose bump, and followers urged legal action. - Bored Panda

LLMs can unmask burner accounts at scale - Ars Technica

sorry all you 'Pierre Delectos'

Colombia's Supreme Court denied a cassation appeal after citing Winston AI detector's 7% human score, but attorneys later ran the court's ruling through the same tool and found it scored 93% AI-generated. - Decrypt

Follow the latest AI headlines via druce.ai on Bluesky