we’re back! had to skip yesterday for reasons

Mythos and Legends: Anthropic's Mythos AI allegedly breached NSA and Cyber Command classified systems in hours - The New York Times

Squid game: Mythos discovers 'Squidbleed,' a memory leak that's gone undetected since Clinton era - theregister

OpenAI's GPT-5.5-Cyber scores 85.6% on UC Berkeley's CyberGym benchmark, edging out Anthropic's Mythos model, which the U.S. government banned for foreign nationals - Decrypt

Zero-day of reckoning: China's 360 Security unveiled Tulongfeng, a self-described Mythos-equivalent AI for discovering software vulnerabilities, and Yitianzhen for automated cyber defense - Reuters

Anthropic CEO Dario Amodei sidelined in White House export control talks, company turning instead to cofounder Tom Brown - WIRED

Slack to the future: Anthropic launches Claude Tag, a Slack-native collaborative AI agent for Enterprise and Team customers that works asynchronously across channels. - Anthropic

Anthropic's Claude Tag brings an always-on AI agent to Slack channels with isolated permissions and admin spend controls - ZDNET

Distill Bill (or Dario): Anthropic accused Alibaba of conducting the largest known distillation attack on its Claude AI model using nearly 29 million exchanges. - The Wall Street Journal

Spicy silicon: OpenAI and Broadcom unveil Jalapeño, a custom LLM inference chip promising 50% lower costs than Nvidia GPUs - OpenAI

Broadcom and OpenAI unveil Jalapeno, a custom AI accelerator chip designed to lower token costs and boost computing efficiency. - CNBC

Codex Maximus: OpenAI's Codex agent now accounts for 99.8% of weekly output tokens at OpenAI, with non-developer adoption rising 137x since August 2025. - OpenAI

Set it…and forget it: ChatGPT's new Scheduled Tasks lets users set recurring briefings and reminders without manual prompting - TechRadar

OpenAI hired DeSantis's former general counsel to defend against Florida's lawsuit alleging ChatGPT endangers children. - Politico

OpenAI co-founded the Appia Foundation to develop open, interoperable AI safety assessment standards hosted by the Linux Foundation. - OpenAI

OpenAI, Anthropic, Amazon, Microsoft and others launch Raise Us, a $500 million nonprofit led by Gina Raimondo to prepare US workers for AI displacement. - The New York Times

Absolutely fabless: IBM unveiled a SOTA chip demonstrator from its labs, but they don't have fabs that can make it soon, would need to license it or partner with e.g. TSMC - Ars Technica

TSMC call me maybe

DRAM Queen SK Hynix, the leading AI memory chip supplier, seeks $29 billion via a US stock listing as high-bandwidth memory demand surges - Reuters

seems a bit bearish that they want to sell so much stock when they are just about the highest market cap stock in Korea

Qualcomm acquires AI software company Modular for $3.9 billion to speed AI deployment across diverse hardware - The Wall Street Journal

Micron's strong quarterly forecast drove Nasdaq 100 futures up 2.1% as AI chip demand optimism lifted US tech stocks. - Bloomberg

Money Context Protocol: AI industry spends millions in New York Democratic primary, signaling broad midterm political spending ahead - The Washington Post

AI industry super PACs spent $27 million on a New York primary over AI accountability legislation; the anti-regulation side narrowly won - Bloomberg

Tech workers and unions launch Guardrails Alliance, a $15 million super PAC to counter pro-deregulation AI money in midterms - The New York Times

America's AI build-out is driving new inflation pressures on consumer goods and electricity even as trade war impacts ease. - The Wall Street Journal

Bipartisan bill would require tech companies to pay for grid upgrades triggered by AI data centers drawing 100+ MW - CNBC

Canada's AtkinsRealis files with the US NRC to certify its Candu 6 nuclear reactor design for AI data center power deals - Bloomberg

Tesla, Sunrun, and Renew Home launch a 16-gigawatt virtual power platform to supply AI data centers with aggregated home batteries and thermostats - The New York Times

Silicon smugglers: US export crackdown doubles black-market Nvidia GPU prices in China as illicit supply dries up - Financial Times

China reclaims the Top500 supercomputer crown for the first time since 2017 with a CPU-only system - The New York Times

Netherlands lobbies US to drop proposed chip export controls that would restrict ASML's older immersion DUV machine sales to China. - Bloomberg

Trump administration explores three mechanisms to give the US government a stake in leading AI companies. - Yahoo

Global AI revenue hit $25 billion in Q1 2026, exceeding $21 billion in data-center depreciation costs for the second consecutive quarter. - Bloomberg

Cerebras Systems forecasts core operating margins of negative 28% to 32% for 2026 amid AI buildout costs. - The Wall Street Journal

Meta plans to replace human content moderators with large language models - Financial Times

Zuck around and find out: White House presses Meta to submit AI models for voluntary government security review; Meta says it hopes to sign soon - TechRadar

Meta launches three new smart glasses starting at $299, including the Kylie Jenner Starfire edition at $399 with an AI voice. - WIRED

Ukraine drone analyst says AI that autonomously selects and strikes targets is the near future for attack swarms. - Business Insider

Defense tech VC hit $12.3B in H1 2026, surpassing all of 2025 as drones and battlefield AI go mainstream - Financial Times

Open tech job postings rose nearly 14% year-to-date in 2026, with hardware engineering roles up 52%, countering AI displacement fears. - Business Insider

MIT researchers estimate 350,000 Greater Boston jobs are exposed to AI, with market research analysts and customer service roles most at risk. - BostonGlobe.com

Journalist Sarah O'Connor argues AI's biggest workplace threat is degrading job quality rather than eliminating jobs. - Bloomberg

Study of 26,811 students finds AI use raises homework scores but cuts exam performance by up to 24%. - Inc.

AI rollouts are burning out top performers tasked with leading adoption, raising retention risks for employers. - Fortune

Retire prompt-ly:Gen X and Baby Boomer workers are learning AI to extend their careers just long enough to reach retirement. - Business Insider

Welcome to the promptocene: World Economic Forum analysis identifies "AI work architect" and "AI steward" as emerging roles as AI handles more routine execution. - Allwork.Space

AI researchers estimate a 10–50% chance of catastrophe and call for a US-China pact before AI can rewrite its own code - NRC

Tech leaders Vint Cerf, David Bray, and Cheryl Strauss Einhorn outline frameworks for trust and accountability in AI governance. - ZDNET

London's Met Police gets 12-month Palantir extension after mayor blocked £50m contract for procurement rule breach - the Guardian

Knicks leak elf-on-a-shelf data and you’re in it: MSG compiled a dossier on activists who publicly criticized its facial recognition system, revealed in a 45 GB data breach posted online. - 404 Media

Nvidia launches Halos for Robotics, its first full safety framework for physical AI systems, with Agility Robotics as the first adopter. - SiliconANGLE

ByteDance unveiled Seedance 2.5, which generates 30-second 4K videos from a single prompt and accepts up to 50 reference inputs. - CNET

OpenAI, Google, and Anthropic are deploying internal AI agents for multistep tasks, with employees repositioned as reviewers - The Wall Street Journal

Gartner warns AI coding agents may soon cost more per month than the developers using them, with no spending guardrails - theregister

The tokenmaxxing era was brief. We now appear to be entering the era of token rationing. - TechCrunch

AWS extends Lambda runtime to 8 hours with new MicroVMs, making serverless computing viable for long-running AI agent tasks - theregister

New boutique hedge funds say AI agents let them compete with established giants in macro and bond investing - Bloomberg

Modal launches Auto Endpoints for self-serve, production-grade LLM inference with full code and metrics transparency. - Modal

A GitHub senior director built 40 automations in the GitHub Copilot app to handle operational overhead across calendar, email, and code repositories. - The GitHub Blog

Medicare's WISeR AI prior-authorization program in Washington state is causing treatment delays and denials, drawing calls from doctors and lawmakers to overhaul it. - The Seattle Times

Google AI researchers are leaving for Anthropic and OpenAI, driven largely by the prospect of pre-IPO equity gains. - Business Insider

AI weather models matched NOAA forecasts for Hurricane Melissa, the strongest Category 5 on record, suggesting they can generalize beyond training data - Bloomberg

Crispr co-inventor Jennifer Doudna is skeptical AI will replace human creativity in drug discovery, challenging industry hype around AI's research potential - Bloomberg

AI-generated mass emails are flooding local government comment processes, raising alarms about fake public input on energy policy - Bloomberg

Washington Post test: ChatGPT gave left-leaning answers to 80% of political questions; Gemini was balanced 90%+ of the time - The Washington Post

RightCapital launches Iris AI agent for financial planners, offering real-time plan review, inconsistency detection, and retirement simulations. - Wealth Management

FIFA is providing all 2026 World Cup teams a Lenovo-powered AI agent to narrow the data analytics gap between wealthy and smaller nations. - WIRED

Survey finds 93% of organizations have experienced infrastructure incidents attributable to AI deployments. - theregister

Krea releases two open-weight 12B-parameter image generation models capable of producing 2K images in 2 seconds. - VentureBeat

ElevenLabs released a 13-hour AI-narrated Odyssey audiobook using Michael Caine's licensed voice ahead of Nolan's film. - Variety

Microsoft used Copilot AI to link two previously unconnected malware tools, Amadey and StealC, then took both down with a single civil lawsuit - Bloomberg

If you've never seen this Burnistoun clip, enjoy: - YouTube

Follow the latest AI headlines via druce.ai on Bluesky