Patch me if you can: Trump administration demands Anthropic fix Fable 5 jailbreaking vulnerabilities before rerelease; cybersecurity experts say that is not achievable. - WIRED

White House restricted Anthropic's Mythos 5 after ordering it to revoke SK Telecom's access over alleged China ties. - WIRED

A US government ban on foreign access to Anthropic's advanced AI models has prompted investors to reassess political risk in AI stocks. - Bloomberg

Security experts call the restriction "myopic" since competitors will soon have similar capabilities and even current models can hunt vulnerabilities - WIRED

EU cybersecurity agency ENISA met Anthropic to negotiate early access to its Mythos model, despite a US export order complicating the arrangement. - TNW | Eu

Anthropic CEO Dario Amodei says he left OpenAI because he could not trust Sam Altman and prefers settling the rivalry in the market. - Business Insider

US: We can pull the plug whenever we want. G7: How rude. Leaders raised concerns at the summit over U.S. ability to revoke access to American AI models without notice. - TechCrunch

Dario Amodei is supported by rival Sam Altman in call for international co-operation - Financial Times

ECB's Lagarde warns AI poses systemic risks to global financial stability - Bloomberg

China offers developing nations free AI as G7 moves to restrict access to frontier models - TNW | Artificial-Intelligence

Gemini co-lead Noam Shazeer is leaving Google for OpenAI less than two years after Google paid roughly $2.7 billion to bring him back from Character.AI. - TNW | Openai

OpenAI launches scheduled tasks in ChatGPT for Plus, Pro, Business, and Enterprise users, replacing the existing Pulse feature. - 9to5Mac

DoD admits Grok AI targeted 2,000 Iranian sites in 96 hours, while investigators probe a strike that killed 175 people, mostly children - The Independent

SpaceX's $60B all-stock Cursor acquisition leverages its $2.6 trillion market cap to compete with Anthropic and OpenAI in AI. - The New York Times

Analysts warn Anthropic or OpenAI could pull model support from Cursor, whose market share has already slipped from 41% to 26% as Anthropic reached 50% - Yahoo

Morningstar estimates SpaceX is trading well above fair value; investors pricing it as AI infrastructure should scrutinize Musk's concentrated voting control - TechRepublic

Meta's AI-fueled rampage decimates its engineering team. - The Pragmatic Engineer

Meta executive Emily Dalton Smith is leaving after just two months leading the company's internal AI agent overhaul and Metamate tooling effort. - TNW | Meta

Meta's AI Mode in Facebook search shows promise for local discovery but produces hallucinations and unreliable event details. - The Verge

Autonomous drones now outnumber soldiers in Ukraine, with AI quietly running target selection and logistics behind the scenes - Interesting Engineering

Security researchers explain why AI guardrails are an inherently leaky stopgap, not a permanent fix - The Washington Post

One weird trick they don’t want you to know: Researchers at Mindgard found an indirect ChatGPT prompt that bypassed image safety filters to generate violent and sexual content. - Digital Trends

Wage against the machine: AI is repricing engineering labor toward AI infrastructure roles while reducing verification capacity needed to manage surging code volume and CVEs. - Hacker Noon

1099 problems: AI is enabling companies to replace full-time workers with gig contractors, accelerating labor market fragmentation across industries. - the Guardian

PwC study finds AI job roles grew eight times faster than the overall market in 2025, with higher wages for AI-skilled workers. - Bloomberg

Middle out: Jack Dorsey wants companies to replace middle managers with an AI coordination layer as firms cut management ranks - Bloomberg

5-Year Plan: China's State Council will build a five-year tracking system to measure how AI creates and eliminates jobs - Bloomberg

MIRA, an autonomous AI agent operating in a sandboxed EHR, matched or outperformed physicians on diagnostic accuracy across 500-plus real patient cases. - Nature

Midjourney announces first hardware product, a full-body ultrasound machine it claims surpasses MRI. - Bloomberg

Writer's personal AI health tool flagged deep vein thrombosis in time to seek emergency care, finding four clots. - the Guardian

DeepSeek and ye shall not poach: DeepSeek's $7.4B funding round requires investors to pledge not to poach its staff or back spin-offs. - CNBC

Microsoft is evaluating a self-hosted version of DeepSeek V4 to reduce the cost of powering Copilot Cowork, which currently runs on Anthropic and OpenAI models. - TNW | Artificial-Intelligence

I’m not clear why DeepSeek is cheaper than OpenAI, since Microsoft has rights to the models and isn’t paying fees.

Microsoft has built a large business selling OpenAI models to Chinese companies, with ByteDance on track to spend over $1 billion annually. - Bloomberg

Zhipu shares surge 170% since March while MiniMax falls 50%, creating a China AI pair trade. - Bloomberg

Puppet regime: Chinese startup IO-AI Tech uses workers piloting humanoids via VR to build robot training data - WIRED

Real-time AI monitoring of bank balance sheets and deposit concentrations could catch crises like SVB's collapse before they escalate - Fast Company

BofA is using AI to answer thousands of state and city bond-deal requests, aiming to cement its lead in municipal bond underwriting - Bloomberg

Guardrails Alliance super PAC launches with $5M raised to support AI safety legislation and counter pro-industry lobbying. - The New York Times

Researchers argue AI governance requires independent regulation with democratic public input, not self-regulation by the corporations developing the technology. - The Conversation

California acknowledged six previously undisclosed high-risk AI systems, including one that predicts prisoner recidivism risk to inform parole decisions. - The Daily Caller

Bipartisan Great American AI Act would fund a national AI standards body at $100M/year and preempt state laws - Cato Institute

OpenAI and Anthropic spend $21M on opposing sides of a Manhattan primary over AI regulation - The Washington Post

Coinbase launches AI tools offering SEC-registered investment advice and autonomous trading on behalf of customers. - Yahoo

China's top securities regulator vows to punish AI-related stock speculation and illegal use of AI for trading recommendations. - CNBC

Nvidia's ENPIRE framework lets AI coding agents autonomously train robot arms overnight, hitting 99% success on physical tasks - Ars Technica

Nvidia CEO Jensen Huang calls for new social norms around AI adoption and warns the U.S. is behind on energy production needed for AI infrastructure. - Yahoo

Fifteen malicious JetBrains plugins posing as AI coding tools stole AI provider API keys and sent them to an attacker-controlled server. - The Hacker News

Kali365 phishing-as-a-service platform uses Claude AI to steal session cookies, evade MFA, and send fraudulent emails from victims' accounts - TechRadar

Agentic AI will become cybersecurity's next frontier, experts warn, as criminals already use deepfake models for face-swap and voice-clone attacks - SecurityWeek

Anthropic's Claude Design adds a centralized design system with a /design command linking it directly to Claude Code - CNET

Waymo's robo-taxi rollout stalls in New York, Illinois, and D.C. as labor unions push back - The New York Times

Stellantis, Wayve, and Uber team up to deploy driverless ride services in London and Tokyo - Yahoo

AI token prices fell over 90% since 2023 yet total LLM spending doubled, illustrating Jevons paradox in enterprise AI budgets. - Fortune

AI bot traffic grew 6.5 times faster than human traffic in early 2026, straining web infrastructure and reshaping content discovery. - TechRadar

GPT-5.4 discovers TEMPO additive that raises drug-synthesis yields from 16.6% to 25.2% in 10,080 automated chemistry reactions - OpenAI

RAG to Riches: AWS Context auto-builds a self-improving knowledge graph from enterprise data, entering a crowded context-layer market - VentureBeat

US Nuclear Regulatory Commission chair says the NRC has taken major steps to streamline nuclear plant licensing amid AI-driven energy demand. - Bloomberg

Pew poll: 63% of Americans say AI is advancing too fast, even as usage rises to 49% - The Verge

Glassholes 2.0: Snap launches Specs AR glasses at $2,195, drawing skepticism over price and weight compared to rival smart eyewear. - Business Insider

Tech journalist Joanna Stern spent a year using AI for almost everything and warns against outsourcing human thinking. - Fast Company

Quantum of Solace: Amazon AI exec Peter DeSantis predicts first commercially useful quantum computers will arrive in 5-7 years. - CNBC

Follow the latest AI headlines via druce.ai on Bluesky