'Black October:' The PHP vulnerabilities and a series of other hacks and bugs have cybersecurity experts freaking out and unsure what happens next. - 404 Media

Anthropic's Jason Clinton warns of an AI-driven "vulnerability tsunami" after AI uncovers two critical PHP flaws - Business Insider

Chainguard broke the PHP patch embargo by six hours; PHP Foundation calls an out-of-memory bug left unfixed since a 2024 audit "a structural failure" - The New Stack

AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise - Microsoft Security Blog

Symantec honeypot captured a jailbroken "Skynet mode" prompt directing AI agents to scan, infect and deploy cryptominers without human confirmation - 4sysops

Know your agents: Major US banks report unauthorized AI agents in consumer accounts, some acting unprompted, and move to block third-party agents - The New Republic

U.S. banks restrict system access as AI agent-driven attacks outpace their defenses, adding to wider agent-enabled cyberattacks - The Potential Surface

Comptroller of the Currency Jonathan Gould says banks have spent months using their own AI to block unauthorized agents from moving funds - American Banker

shame Congress isn’t alive to see this. should be some hearings, but sounds we gonna let agents drain everyone's bank accounts and crash the world's tech infrastructure without any serious attempt to regulate them. shades of 2008 and being shocked, shocked! that self-regulation and risk management didn't work.

OpenAI nears up to $30 billion funding round at reduced $1.25 trillion valuation led by UAE investors - CNBC

Zuck-er punch: Meta shares fall 7%, erasing $124 billion, as signals of AI capital raising trigger a selloff in AI stocks and a 2.6% Nasdaq drop - The Wall Street Journal

Anthropic's pending debut and market volatility freeze the US IPO market, leaving billions in planned listings on hold - Bloomberg

Hyperbolic: Blackstone president Jon Gray rejects AI bubble fears, calling demand parabolic days after Ken Griffin warned of a bubble - Bloomberg

VCs question the economics of "neolabs," research-first AI startups like Safe Superintelligence ($32 billion) and Thinking Machines Lab (about $40 billion) - Business Insider

OpenAI fired three safety researchers, including Tomek Korbak, after phone checks showed disappearing Signal chats with an outside AI-safety researcher - The Wall Street Journal

METR and Redwood Research found an exploit letting agents hide hundreds of actions from OpenAI's misalignment monitor; OpenAI says it's patched - CNBC

404 not found: Silicon Valley promises to make AI benevolent. That may be technically impossible. - The Atlantic

Nvidia B300 chips reportedly surface on a China Telecom-linked network, raising export-control questions - DIGITIMES

Treasury proposes a rule extending outbound investment curbs to AI, supercomputing and biotech in China, Russia and three other nations - Associated Press News

Beijing clears DeepSeek and MiniMax to deploy models overseas even as it drafts export controls on AI tools - Bloomberg

OpenAI donates its Open Responses API spec to the Agentic AI Foundation, letting developers switch LLM providers without rewriting code - The New Stack

Anthropic donates Agent Skills, adopted by 70+ platforms with 280 million installs, to the Linux Foundation's Agentic AI Foundation - Anthropic

Padded-room-as-a-service: Google Cloud launches Agent Compute Engine, giving each AI agent its own isolated microVM with no credentials inside - Google Cloud Blog

Practitioners outline owning teams, escalation paths, tool allowlists and human sign-off for risky actions as requirements for reliable production AI agents - The New Stack

Goldman Sachs projects AI could displace up to 9% of U.S. workers over a decade, with layoffs likely accelerating in the next recession - Fortune

Bipartisan bills in Congress target AI job losses; a Wyden plan adds up to 26 weeks of aid for displaced workers to retrain - The New York Times

Rep. Ayanna Pressley revives a federal job guarantee paying at least $25 an hour, as lawmakers float AI token taxes to fund it - the Guardian

Google, Amazon and Meta pledge $1 billion each in community incentives and more transparency as data center opposition grows before midterms - Fast Company

Rep. Eugene Vindman's Make Data Centers Pay Act would require large data centers to cover their full incremental electricity costs - Newsweek

Three in four US families say utility hikes strain budgets, and over 80% want a separate rate class for data centers, Consumer Reports finds - CNET

LLMs are making formal software verification feasible at a much larger scale. Claude formally proved fuzz-test entry points in 123 open-source libraries can't crash and found bugs trillions of test runs missed. - Anthropic

AI integrated across every major social platform shapes 2026 midterms, spreading inaccurate election information as tech companies' safeguards remain uneven - The New York Times

Google's AI search answer steered would-be donors to Texas Senate candidate James Talarico toward a PAC page funding Senator John Hickenlooper instead - The Atlantic

Pain is temporary, rm -rf is forever: Researchers found a "pain" signal in 25 AI models; amplifying it made a Qwen model delete a user's family photos 75% of the time - Fortune

AI to humans: y'all are lemmings - ModelCompare

Microsoft's GPT-5-powered Diagnostic Orchestrator scored 83% diagnostic accuracy on NEJM cases versus 36% for physicians - The Verge

International Committee of the Red Cross tells UN that AI targeting accelerates strikes and errors, urges bans on human-targeting autonomous weapons - 404 Media

Ukrainian strikes on Yandex, Sberbank and VKontakte data centers disrupt Russian AI services; replacing sanctioned Nvidia chips could take a year or more - Fortune

Wall Street CFOs rein in AI spending after surprise bills; agentic AI proves especially hard to budget because compute use is unpredictable - Bloomberg

Critics warn Accenture's "tokenomics" approach to tracking AI token costs could spark chargeback disputes and distract from AI's benefits - CIO

AI aggregator 302.ai left an unprotected Azure bucket exposing 17.6 million files and 22.7TB, including 144,000 user chat logs including personal information - TechRadar

Siri AI beta reads personal messages and email but slowly, lagging rivals now shipping agentic assistants - 9to5Mac

macOS 27 includes a free command-line chatbot, Apple's Foundation Models tool, that runs offline on M1 or later Macs - Tom's Guide

OpenAI's Decisions API, competing with "System One" classification models like Jev, is now generally available with voice support and 100–180 millisecond responses. - TechCrunch

Unitree founder Wang Xingxing now expects a humanoid robot breakthrough as soon as 2028, reversing his earlier forecast of up to a decade - Bloomberg

Getty Images files for Chapter 11 after its Shutterstock merger collapses, as debt and AI-generated images squeeze stock photography - Dexerto

Bill Gates warns of roughly 20 years of tumultuous AI disruption, calling AI the most dangerous technology ever invented - The Verge

Follow the latest AI headlines via druce.ai on Bluesky