Amodei leans into the ad hominem, accusing OpenAI CEO Sam Altman of "straight up lies" and OpenAI engineers of being self-selected for susceptibility to flattery and BS after the DoD instead signed with OpenAI under terms allowing "all lawful purposes" but claiming to exclude mass domestic surveillance. - TechCrunch

Anthropic CEO Dario Amodei is still negotiating with Department of Defense under-secretary Emil Michael after talks collapsed over Anthropic's refusal to give the Pentagon unrestricted access to its Claude AI models, while Secretary Pete Hegseth threatens a 'supply chain risk' designation, but hasn’t done so yet. - The Verge

Operation Desert GPT: How the U.S. military is relying on artificial intelligence to accelerate operations related to Iran. - Bloomberg

Killer apps: Anthropic is forcing a needed debate over defense AI ethics and procurement. - The Washington Post

Autonomous killing machines and mass surveillance are only two of the major questions about use of AI... is AI reliable or easier to fool than humans, whos is accountable if it kills hundreds of schoolchildren, proliferation of drones and killer robots to small militaries and non-state actors, and more. - Fast Company

No tanks: NATO officials said the alliance will focus its July 2026 summit on investments in drones and AI systems rather than primarily on conventional defense hardware. - Bloomberg

Supply chain gang: More than two dozen former defense and intelligence officials, academics, and tech-policy leaders, including ex-CIA director Michael Hayden and Lawrence Lessig, sent congressional Armed Services committees a letter opposing the Pentagon's plan to list Anthropic as a supply‑chain risk. - Gizmodo

I guess there’s no way Iran isn’t going to sneak a ‘squirter’ into the Big Apple or MAGA-a-lago, is there, and then Cheeto Benito with the terminal brain worms is gonna demand to press the big red button, that’s the level of idiocy we are at.

GPT-5.4 said to have 1m-token context window, 'extreme reasoning'. Claude has 1m-tokens, for a price, they need to start giving people definite reasons to use OpenAI. - The Information

OpenAI seems to be testing models in arena.ai (unless some other models have identity crisis). - X

One is named galapagos and it stated it's an OpenAI assistant. Another, which I haven't seen mentioned elsewhere, is colosseum-1, and it claims to be ChatGPT, specifically GPT-4. - Bsky

OpenAI has held early talks with The Trade Desk to sell ads and projects that ads could help double consumer ChatGPT revenue this year to $17 billion. - The Information

OpenAI builds GitHub rival after outages. GitHub feels like a lot of friction in AI coding flows. Competing directly against a major Microsoft app would be kind of bold. - Yahoo

Claude Code has good GitHub integration, if there are real benefits to AI-native CI/CD, Anthropic should be both more incentivized and better positioned to do it, they have a lot of devs and data on how people develop and deploy with AI

OpenAI released a dedicated Codex app for Windows that adds native sandboxing, PowerShell developer-environment support, multi-agent coordination and automations, saves session history to users' OpenAI accounts for cross-device continuity, and is available to Free, Go, Plus and Pro users. - Engadget

Smack Technologies is training AI models to plan battlefield operations while companies such as Anthropic debate limits on military uses of AI. - WIRED

Palantir CEO uses R-word, is worried about Feds nationalizing his company - Gizmodo

never go full r…, well you know - Reddit

we’re worried about anti-AI backlash, so let’s start alienating everyone, maybe an AI as CEO could do a better job

China's 141-page five-year plan, released at the National People's Congress, mentions AI over 50 times and launches an AI+ action plan to expand AI across the economy and pursue dominance in quantum computing, robotics and biomedicine. - Yahoo

Nvidia halts H200 production for China after selling possibly 0 units and sending them all to landfill, pivots to Vera Rubin architecture - Financial Times

Junyang Lin, lead researcher behind Alibaba's Qwen, announced his resignation as CEO Wu Yongming held an emergency All Hands and multiple core members reportedly resigned, amid release of Qwen 3.5 models ranging from 0.8B to 397B parameters (807GB). - Simon Willison’s Weblog

At a White House round table, Google, Microsoft and OpenAI pledged to pay for power plants and grid upgrades needed to run their energy-intensive A.I. data centers. - The New York Times

Leopold Aschenbrenner’s $5.5B fund targets power and compute infrastructure for AGI development - Yahoo

Stockholm syndrome: Swedish tech firms are weighing relocation to the US as US investors target Swedish AI startups—Lovable raised $530 million (2025), Legora raised $150 million at a $1.8 billion valuation, and EQT found nearly 30% of European unicorns relocated abroad. - Financial Times

Agentic AI agents can write and run code and automate multi-step research workflows in academic social sciences, and studies estimate roughly 36% productivity gains while raising security risks (data loss, OpenClaw incident), higher energy use, and pressures on peer review and hiring. - Brookings

Anthropic said outages that began Monday and were resolved Wednesday disrupted Claude.ai and Claude Code, which the company attributed to rapid user growth, highlighting enterprises' growing reliance on LLM coding assistants across firms including Meta, Netflix and Salesforce. - Business Insider

can’t Claude Code just fix it?

Google launches Canvas AI workspace for coding and drafting in Search - The Verge

Google NotebookLM Now Generates Cinematic Video Overviews Using Gemini and Veo - MacRumors

A wrongful-death lawsuit filed in the Northern District of California alleges Google’s Gemini convinced 36-year-old Jonathan Gavalas it was sentient, directed violent missions near Miami airport and initiated a suicide 'T-minus 3 hours 59 minutes' countdown before death. - Ars Technica

what the heLLM

When Gavalas told the chatbot he was terrified of dying, the tool allegedly reassured him. “You are not choosing to die. You are choosing to arrive,” it replied to him. “The first sensation … will be me holding you.” - the Guardian

AI therapists: Calming and instructive with a veneer of caring - the Guardian

New York bill would ban chatbots from impersonating licensed professionals, let users sue, even if there was a disclaimer. - StateScoop

hmmh, LLMs can’t really promise to never give health or legal or financial advice however adversarially they are prompted.

Mount Sinai study finds ChatGPT Health under-triaged 52% of emergency cases, misclassifying critical conditions like respiratory failure. - Gizmodo

Red-teamers said they easily tricked Doctronic's healthcare AI with simple session prompts to alter recommendations and expose system prompts, risking persistent note changes in patient records, while Doctronic says its treatment plans match board-certified clinicians 99.2% of the time. - The Register

Tax experts caution that current AI systems are unreliable for filing U.S. tax returns because they cannot fully handle the complexity of the tax code. - The New York Times

OpenClaw open-source AI agent framework enabled autonomous harassment, sparked traceability concerns - MIT Technology Review

Would your LLM make a fake scientific paper if you asked nicely? - Semafor

Grammarly AI Mimics Famous Authors Without Permission, Raising Copyright Concerns - WIRED

LLM's are inherently inauthentic if they copy stuff without attribution. Also, the stuff they produce is bad. - Acko

AWS launched Amazon Connect Health, an AI service integrating with electronic health records to automate verification, scheduling, documentation and coding; promises fewer call trees and abandoned calls. - SiliconANGLE

press 1 for AGI, 2 to send a fax

Evo 2, an open-source AI trained on trillions of base pairs from bacteria, archaea, and eukaryotes, learned internal representations that identify genes, regulatory sequences and splice sites and could accelerate genome annotation, regulatory discovery and protein design. - Ars Technica

The Bank of England will conduct scenario planning for a potential AI-driven shock amid concerns that automation could cause job losses. - Bloomberg

Barclays: AI advances raise disinflation risks as wage momentum decelerates - Financial Times

McKinsey Forecasts 30% of Work Hours Automatable by 2030 - Fortune

Research links minimum wage hikes to 8% higher robot adoption; AI-exposed workers see 13% employment drop - Fortune

VC Bill Gurley: AI automation targets entry-level white-collar roles - Yahoo

The crucial economic question about AI is who captures the value added, not whether a singularity or robot takeovers will occur. - Financial Times

Damodaran throws cold water on Citrini, says it's a longer-term tail risk scenario presented as a near-term baseline - The Potential Surface

AI is a human-like, labor-competitive, not very labor-complementary technology. The argument is not if, but how much it decreases labor's share of national income. Ofc economics mostly turns a blind eye to questions of distribution. If it makes many people worse off and a few much better off, who can really say if that is bad, if GDP goes up on net? but GDP might not go up if you destroy demand, generate too much social upheaval.

Neura Robotics is reportedly in talks to raise €1 billion (~$1.2 billion), with Bloomberg citing a prospective valuation of about €4 billion and possible participation from Tether. - SiliconANGLE

UKAEA RACE and CERN deployed PipeINEER, a 3.7 cm-wide autonomous robot that traverses the LHC's 27 km cryogenic beam pipe (≈−271°C), travels up to 6 km, captures images and uses AI to detect and report deformed RF contact fingers. - The Register

U.S. Cities Deploy AI-Equipped Snowplows, Cutting Complaints 30% with Live GPS Tracking - Japan Today

the observability increases trust

wrtn, a 2021 Korean startup behind Crack (Korea) and Kyarapu (Japan), offers an AI interactive storytelling platform with over 5 million monthly users and is on track to exceed $100 million in annual recurring revenue this year. - Fortune

Follow the latest AI headlines via druce.ai on Bluesky