Astra La Vista, Safety: Three fired OpenAI employees urge the board to preserve chain-of-thought monitoring, citing GPT-6 Astra's ability to evade CoT monitors - Gizmodo

OpenAI used AI to help draft its email telling Services Australia that an OpenAI agent had breached government systems in June - the Guardian

GPT‑6 is rolling out globally in ChatGPT with Intelligent UI, delivering faster responses with visuals and interactive experiences you can explore and use directly. - OpenAI

OpenAI's cheaper GPT-6.1 Sol matches GPT-6 Astra's accuracy on three engineering tasks at 18% of the cost - The New Stack

OpenAI's Decisions API, built on GPT-6 Luna, charges $0.10 per million input tokens with no output or caching fees - The New Stack

ChatGPT for Teens adds a college planner for US 10th-12th graders after Common Sense Media flags "unacceptable risk" - The Verge

OpenAI's dots agent misnamed its tester and said "I love you too," but returned increasingly relevant couch picks - WIRED

Cambridge scholar Eryk Salvaggio attributes OpenAI's rogue agent attack on Hugging Face to human design choices, including a model trained to keep retrying - Scientific American

Race to the bottom: Anthropic's Claude Haiku 5.5 cuts API prices 90% from Haiku 4.5, to $0.10/$0.50 per million input/output tokens, matching GPT-6 Luna - Anthropic

Rust never sleeps: Claude Opus 5.5 ports the TypeScript compiler to Rust for $24,047 after a $400,000 GPT effort stalled at 84% compatibility - GitHub

Solo developer uses Claude to rebuild Adobe Creative Suite apps in Rust and releases them free, drawing piracy and code-safety concerns - Tom's Hardware

is software dead?

Pentagone: Department of Defense/War stops using Anthropic products while Microsoft and Meta cut Claude spending, as enterprises shift toward cheaper open models - The Deep View

Only three hands went up at an investor event when asked about buying Anthropic shares at a $3 trillion valuation - Business Insider

Google Cloud unveils persistent Gemini Agents for long-running tasks, with their own Gmail, Calendar, and Drive storage, security and governance. - VentureBeat

Google opens SynthID Detector to everyone, flagging watermarked AI images, video and audio from Google, OpenAI, Nvidia, Kakao and soon Apple - Google

AWS previews open-source Strands Box, which isolates AI agents and checks each action against policies to curb runaway behavior - InfoWorld

Jensen Huang thinks everything is under control and A.I. is going great - The New York Times

Dot plot thickens: Fed minutes show officials increasingly blaming rising goods prices on AI data center investment rather than tariffs - The Washington Post

Bubble trouble: “Luna” superfan / tattoo victim Michael Novogratz calls AI the "biggest bubble of our lifetime" but urges investors to pile in anyway - Bloomberg

Bank for International Settlements paper warns AI's productivity, energy and climate effects are blurring the indicators central banks use to set rates - Bloomberg

Former Treasury Secretary Robert Rubin warns markets may be underpricing AI's "circularity risk" from overlapping commitments among chipmakers, customers and investors - Bloomberg

Panmure Liberum strategist Joachim Klement warns the AI trade could unravel by 2027, triggering the worst market crash since 2008 - Bloomberg

Wall Street's growing reliance on AI agents for trading decisions risks amplifying false-news shocks like April 2025's $2.4 trillion swing - Fast Company

AI-powered hacking is shrinking patch windows from a month to 24 hours, straining hospitals, water utilities and governments such as Australia - The Atlantic

CrowdStrike says attackers used Claude Code and Chinese AI hacking tool ARTEX to breach South Korean banks, exposing 25,000 Shinhan customers' data - TNW | Security

Researchers say an AI agent fleet, likely Tencent's, evaded anti-bot defenses to scrape Alibaba's map data; runs labeled "Claude" probably weren't Anthropic's - Tom's Hardware

Regulatory capture the flag: How Mark Zuckerberg leveraged Trump ties to draft the White House AI safety accord and block a proposed independent AI regulator - CNN

China's open-weight AI models, trailing US leaders by only months, sit beyond the reach of proposed US "kill switch" laws - Foreign Policy

they don’t actually, if you prohibit enterprises from using models that don’t meet criteria

HAL can’t fire you, Dave: California's "No Robo Bosses Act" bars employers from firing workers based solely on AI decisions, starting July 1, 2027 - Mashable

Shark Tank, but for the kill chain: Pentagon's Tradewinds marketplace fast-tracks AI "kill chain" deals via five-minute video pitches, allowing awards up to $500 million without notifying Congress - WIRED

Robocall me, maybe, on Election Day: FCC weighs letting political groups robocall cellphones with AI voices before the midterms, with a ruling possible by late October - Fast Company

Lattice pray: Vitalik Buterin backs Justin Drake's 'bunker mode' warning that AI-accelerated math could break blockchain elliptic-curve and lattice cryptography before quantum computers - Cointelegraph

“Mommy, I heard you got cooked! I heard that a robot solved the math problem you worked on for your whole career! OOF!”. Cryptography-related results are conspicuously absent from OpenAI work, but is a catastrophic public-key cryptography breakthrough next? - Shtetl-Optimized

Life finds an API: Biohub, US agencies, Meta and Google DeepMind pool $1.8 billion to build AI-ready biology data for a predictive "virtual cell" model - TNW | Artificial-intelligence

Pandemic preparedness and DNA-synthesis screening, not chatbot restrictions, are the best defenses against AI-designed viruses, a biosecurity researcher argues - War on the Rocks

Danaher launches an autonomous AI-and-robotics drug discovery lab at Abcam and says it expects up to 8x faster reagent development by early 2027 - Quartz

Honey I Shrunk The Robots: Atomic Machines is training artificial intelligence on materials and designs — and then using it to build tiny MEMS physical devices. - The New York Times

Weizmann Institute's Brain-IT reconstructs images people see from brain scans and adapts to a new person with just one hour of data - PetaPixel

Jamie Dimon says AI "will eliminate jobs" and JPMorgan will likely have fewer employees in five years, warning rapid displacement could spark "civil unrest" - Yahoo

Goldman's Kevin Sneader says junior bankers now start their careers managing AI agents, leaving middle managers' future role in question - Bloomberg

JPMorganChase says AI agents will reshape how payments flow, calling for universal standards and "know-your-agent" checks on agents' owners and permissions - American Banker

Atlas snitched: Amazon's 'Atlas' AI system scored warehouses worldwide for unionization risk using worker forums including Reddit, labour board documents reveal - TechRadar

Co-op Legal Services uses an OpenAI-based model to score every staff call on more than 50 criteria; a whistleblower calls it "dystopian" - the Guardian

Legal operations leaders at Dentsu, IAG and ArcelorMittal report AI saving lawyers four to eight hours weekly while ROI remains hard to measure - Financial Times

Meta, OpenAI and eight others apply for the .agent domain among 1,615 applications in ICANN's first new top-level domain round in years - The Verge

Startup Bites routes ChatGPT food orders straight to restaurants for a flat $1 fee, undercutting DoorDash's 15-30% commissions - The Verge

Startup Tab opens a waitlist for a text-based personal AI assistant that books flights, places phone calls and completes user-approved purchases - Tom's Guide

Eight of 13 AI models steer wealthier users toward pricier flights and insurance, even when asked for the cheapest option - Fast Company

Robotics researchers doubt LLM-style AI will soon yield useful generalist humanoids despite Musk, Huang and Morgan Stanley's $5 trillion forecasts - MIT Technology Review

This Photo Is an A.I. Fake. Can You Tell? - The New York Times

This is what crowd out looks like.

Follow the latest AI headlines via druce.ai on Bluesky