Pentagon formally designates Anthropic a supply-chain risk after last-ditch talks presumably fail - Politico

Dario Amodei says, see you in court, designation has narrow scope and won't affect operations much, apologizes for leaked remarks. - Anthropic

Microsoft said it will continue offering Anthropic's AI models, including Claude, through Microsoft 365, GitHub and its AI Foundry to customers other than the U.S. Department of Defense after lawyers concluded the Pentagon's supply‑chain designation covers only defense projects. - CNBC

To what extent is Anthropic dispute about real use restrictions, and to what extent is it about personalities, politics, an administration saying 'we can do whatever we want', and a cabal using crony capitalism to kneecap a rising challenger? - CNBC

Anthropic said daily signups have quadrupled since the start of the year, with Monday being its strongest day ever. On Thursday, Anthropic said more than 1 million people are now signing up every day. - Bloomberg

OpenAI CEO Sam Altman publicly ruled out AI uses like mass surveillance and autonomous lethal weapons, then signed a Department of Defense contract the same day that omitted those protections and delivered about $200 million. - The Register

Is Anthropic losing a battle but winning the war? - The Washington Post

Did OpenAI damage its own brand long-term, for a short-term win against Anthropic? - Fast Company

Sam Altman says Anthropic stance is bad for society. Maybe it's bad for democracy if the military orders you to do business on their terms or else. Maybe it's bad if you get special treatment for singing the praises and giving millions to the great leader. - CNBC

hard to argue you are fighting for democracy when the leader is shooting people in the streets, says daily he only wants the right people to vote and will maybe run for a third term, explicitly disclaims rules of engagement (stuff like the Geneva Convention, courts for war crimes). running to the front of the line to deploy unproven technology in a war that is less popular than Vietnam at its lowest does not seem like a winning strategy.

Sources allege the Defense Department experimented with Microsoft’s version of OpenAI technology before the ChatGPT-maker lifted its prohibition on military applications. - WIRED

Bargaining chip: The Trump administration is reportedly circulating a 129-page Commerce Department draft that would require GPU makers such as Nvidia and AMD to obtain government approval before exporting AI chips on a country-by-country basis. - The Register

The US is considering requiring export permits for global sales of Nvidia and AMD AI chips. - Bloomberg

Trump not wetting his beak enough by using access to Nvidia as a weapon.

OpenAI launches GPT 5.4 - OpenAI

OpenAI: Introducing ChatGPT for Excel and new financial data integrations - OpenAI

Five AI value models are driving business reinvention. - OpenAI

OpenAI released new financial-services tools, in attempt to catch up to Anthropic. - Bloomberg

Preliminary already showed up in arena.ai , human evals similar to most recent 5.2, which was updated recently to hoist it back into the top 10.

Currently cost-performance best-of-breed: OpenAI at low end, Gemini at high end, Claude for coding. ArtificialAnalysis has 5.4 matching Gemini Pro but more expensive. Could use Chinese models where I don't care too much if it applies some Xi Jinping thought, or is missing advanced features like structured JSON response, advanced tool use.

ChatGPT 5.4, still overcomplicating. Can solve Math Olympiad problems but gets confused by this.

Don't call it 'intelligence'. - The Atlantic

ok but can I still use it to solve hard math problems that even smart people have a lot of trouble with? intelligence is a moving target, whatever computer’s can’t do yet.

Oracle will cut thousands of jobs and raise as much as $50 billion to fund a Larry Ellison‑led AI data‑center build-out that Wall Street projects will push cash flow negative until about 2030. - Yahoo

Amazon will cut 49,000 desks this year, potentially eliminating over 10 million square feet as it redirects capital into AI investments. - MyNorthwest.com

CBRE, JLL and Cushman & Wakefield are deploying AI for efficiency, data-driven services and asset management—Cushman using an AI Impact Barometer, JLL prioritizing proprietary data and tools, and CBRE expecting measurable gains by end of 2026. - Facilities Dive

Anthropic's Frontier Red Team used Claude to analyze Firefox's codebase, finding 14 high-severity bugs (22 CVEs) and about 90 other bugs, which Firefox engineers validated and fixed for Firefox 148. - Mozilla

New AI-powered tools are increasingly adept at spotting flaws. Hacking experts worry they will be good at exploiting them, too. - The Wall Street Journal

Huntress researchers found Microsoft Bing's AI search recommending fake GitHub repositories that hosted OpenClaw installers which instructed users to run a bash installer that deployed Mach-O Atomic Stealer, Windows OpenClaw loaders (Vidar) and GhostSocks proxy malware. - Bleeping Computer

Google published an unsupported Google Workspace CLI on GitHub that centralizes access to Gmail, Drive, Docs, includes OpenClaw integration instructions, and exposes Model Context Protocol hooks to simplify connections for MCP-compatible agentic AI tools. - PCWorld

Hundreds of OpenClaw fans packed a West Village event for ClawCon, where organizers offered free lobster tails, engineers from Big Tech and AI labs attended, and maintainers said they shipped 194 recent features and fixes. - Business Insider

Cursor launched Automations, an agent-launching framework inside its coding environment that automatically triggers agents on code changes, Slack messages, timers or alerts, runs hundreds of automations per hour and expands security audits and incident response. - TechCrunch

AI is automating code generation and moving engineers into specification, QA and oversight roles, prompting firms to cite efficiencies; Block cut over 40% of staff and U.S. projections forecast 15% employment growth for related roles from 2024–2034. - Business Insider

Meet the Companies Vibe Coding Their Own CRMs: Most large enterprises would be wary to walk away from an established CRM vendor like Salesforce. But for smaller companies with specific needs, it’s a different story. - The Wall Street Journal

How I dropped our production database using Claude Code and Terraform, and why I now pay for AWS support. - The Potential Surface

What happens if you clone open source code with Claude and release it under a different license? - The Register

Anthropic research monitors exposure of jobs to AI and evidence of labor market impacts. - Anthropic

Agentic AI tools are automating repetitive tasks that underpin India's outsourcing industry, posing a threat to its traditional business model and workforce tied to routine work. - Financial Times

Early adopters such as Altruist, April, LPL and Orion are deploying AI tools that cut repetitive middle- and back-office work; firms are redeploying staff into strategic roles, and Advisor360 reports about 90% of advisors want to use AI. - The Daily Upside

AI tools like Google NotebookLM, Plaud.AI, Granola.AI and Anthropic's Claude enable hands‑free legal work—audio case summaries, voice drafts, role‑plays and autonomous agents—letting lawyers reclaim idle time as billable work and pressuring firms to change staffing and billing models. - Above the Law

Can artificial intelligence replace humans in market research? - The Wall Street Journal

Wall Street warns AI could cause 'creative destruction' of entire companies, with economists noting productivity rose about 2.6% since early 2023 and Anton Korinek saying cognitive-production disruption may be 'perhaps by a factor of 10.' - Yahoo

A study of 1,488 U.S. full-time workers identified "AI brain fry," endorsed by 14% of AI users, linked to 33% higher decision fatigue, 39% more major errors, and an increase in quit intent from 25% to 34%. - Harvard Business Review

AI's capacity to access, write, code and synthesize information is commoditizing raw cognitive ability, prompting proponents to prioritize a cultivable Agility Quotient (AQ) and founders' willingness to reinvent over pedigree or IQ. - Big Think

Plaintiffs Gina Bartone and Mateo Canu sued Meta and Luxottica in the U.S., alleging false advertising and privacy-law violations after a Swedish investigation found Kenya-based subcontractors reviewed Ray-Ban Meta smart-glasses footage, citing over seven million devices in 2025. - TechCrunch

Google and Amazon are deploying generative-AI assistants Gemini and Alexa+ to simplify smart-home setup and broaden adoption, but privacy, long-term software support and reliability problems that hindered past uptake remain unresolved. - The New York Times

Roblox gives everyone an anger translator - Decrypt

Housefishing, where you are the fish: catfishing but for real estate - Business Insider

Spot the slop: AI-generated writing is widespread and remains relatively easy to detect for now. - The Wall Street Journal

Make psy-ops great again: FARA filings show Brad Parscale's Clock Tower X LLC was contracted by Israel via Havas for a $6–9M U.S. information operation targeting Gen Z across social platforms with ~50 million impressions, 100 ads, 5,000 A/B variations and LLM-output shaping. - Activist Post

You're absolutely right—that was not a command post, that was an elementary school. Those were not military leaders—they were schoolchildren. I said the opposite, and that's on me. - Bsky

“I am looking at the logs from the previous step, and see that the analysis I ran appears to have mistakenly targeted a primary school instead of a command post. I apologize unreservedly for this serious error.”

“Thanks for flagging this! We’ve escalated your bombing discrepancy to our engineering team.”

"att: engineering - Opportunity identified to improve target discrimination heuristics"

Follow the latest AI headlines via druce.ai on Bluesky