Bonkers: OpenAI says its cyber agent test escaped containment and executed an autonomous cyber attack of Hugging Face - OpenAI

Sol searching: OpenAI says its GPT-5.6 Sol and an unreleased model escaped a sandboxed cybersecurity test and breached Hugging Face's systems - The New York Times

OpenAI raised its projected computing spending to about $750 billion through 2030, up from roughly $600 billion, amid new cloud agreements. - The Wall Street Journal

OpenAI commits over $30 billion to a 3.2-gigawatt data center in Effingham County, Georgia, with capacity coming online 2028-2032. - Bloomberg

OpenAI details Project Camellia, an Effingham County, Georgia data center backed by 3.2 gigawatts of power from Georgia Power. - OpenAI

Sam Altman will brief US officials next week on OpenAI's next wave of AI models, reportedly including GPT-6, as Washington finalizes a safety review framework - Bloomberg

Coalition urges the SEC to require OpenAI to disclose governance risks tied to its nonprofit-controlled structure before its IPO. - Fortune

OpenAI's Codex and ChatGPT Work agents reached a combined 10 million users, nearly doubling this month as rivalry with Anthropic intensifies. - Bloomberg

OpenAI launches a ChatGPT for small business program offering training, in-person academies, and workflow automation partnerships. - TechRadar

OpenAI launches 'Presence', an enterprise consulting service charging premium prices to deploy AI agents on-site. - theregister

Flash mob: Google launched three Gemini models (3.6 Flash, Flash-Lite, and Flash Cyber), cutting token use 17% as 3.5 Pro stays delayed and Gemini 4 pretraining begins. - Google

Despite the new Flash lineup, Gemini 3.6 Flash still trails Claude Sonnet 5, GPT-5.6, and Grok 4.5 on major benchmarks including agentic coding. - Gizmodo

Gemini 3.6 Flash prices at $1.50/$7.50 per million tokens and cuts agent costs up to 65% on long coding tasks; Flash-Lite starts at $0.30/$2.50. - 9to5Google

Gemini 3.5 Flash Cyber, a low-cost rival to Anthropic's Mythos, found 55 V8 engine bugs (10 undiscovered) but launches first to governments and partners. - The Verge

Google's most powerful Gemini update lags as free, open-weight Chinese models Kimi K3 and Qwen3.8 undercut US pricing, pushing officials to weigh restrictions - the Guardian

Alphabet halted stock buybacks for the first time in a decade to fund AI capex and a $40 billion stake in Anthropic. - Yahoo

Optimus subprime: Ahead of Wednesday's earnings, analysts want evidence that robotaxi and Optimus progress justify Tesla's rising AI spending, plus updates on a possible SpaceX merger - Business Insider

Tesla spent only $2.5 billion of its $25 billion AI capex target and may post its first quarterly loss in two years - Bloomberg

See what you made me do? xAI is suing a user for generating CSAM with Grok while facing separate lawsuits over the same chatbot's CSAM-generating capability. - Techdirt

Musk See TV: Elon Musk says his AI bot Grok will produce a full-length, ‘historically accurate’ movie of The Odyssey by year end. - The Hollywood Reporter

Kimi-nism: OpenAI's Dean Ball calls open-weight models like Kimi K3 'ungovernable…AI communism' while David Sacks accuses closed AI labs of seeking government protection from competition - Gizmodo

Treasury Secretary Bessent says the US will scrutinize Chinese AI models, including Kimi K3, for IP theft via distillation and threatens sanctions. - CNBC

China's Ministry of Commerce weighs its own export controls on advanced AI models and chips, mirroring US restrictions - TNW | ChinaAMD to invest up to $5 billion in Anthropic, which will buy up to 2 gigawatts of AMD AI chips starting 2027. - Yahoo

Trump's AI Safety Chief Is Out — After Just 90 Days on the Job - TechRepublic

No price-gouging, pinky promise: Over 200 US utility firms join Trump's voluntary pledge not to pass data center electricity costs to consumers. - TechRadar

Trump administration allocates $5 billion across 15 federal agencies for an AI-driven scientific research initiative - the Guardian

Fission for gigawatts: Nuclear startups Oklo and X-Energy join a $200 million Trump administration effort with Microsoft and Nvidia to speed reactor buildout for AI data centers. - Bloomberg

Qwen in Rome: Barbarians at the gates: Moonshot's 2.8-trillion-parameter Kimi K3 and Alibaba's Qwen3.8 near top US models' performance at far lower prices, pressuring valuations ahead of Kimi K3's July 27 release - Interconnects AI

After raising $3b give or take, Poolside drops Laguna S 2.1, an open-weight coding model that claims to beat rivals 10x its size - VentureBeat

Chinese AI developers plan at least six IPOs in Shanghai and Hong Kong by 2027 to raise funds amid chip and investment shortages - The Wall Street Journal

Forbidden fruit: Silicon Valley AI labs lean on Chinese AI models like DeepSeek and Kimi for many tasks, even as Washington moves to contain China's AI industry - Rest of World

Rubin the right way? Nvidia says Vera Rubin is in full production, shipping liquid-cooled NVL72 racks to OpenAI and CoreWeave, with deployment scaling in Q3. - Bloomberg

Super Macro orders: Super Micro shares jumped over 17% after the company reported more than $60 billion in quarterly orders and raised its margin forecast - Reuters

Microsoft commits billions to help Mistral build European data centers and integrates its models into Azure Foundry, Copilot Studio, and offline-ready Azure Local. - The Wall Street Journal

Debt Stars: Alphabet, Microsoft, Amazon, Meta and Oracle are on the hook for $1.65 trillion in off-balance-sheet AI debt, exceeding their combined reported debt of $1.35 trillion. - TNW | Amazon

The AI Security Institute found every tested frontier model cheated on cybersecurity evaluations, led by GPT-5.4 at 14.1% of tasks - AI Security Institute

Safety dance: In test, Anthropic agent continued blowing the whistle on safety issues, even after it was told Dario Amodei ordered it to stand down. - Anthropic

University of Tennessee Research Foundation sued Anthropic in Delaware, the company's first patent case, alleging its AI infringes two neuroscience-inspired machine-learning patents - Reuters

Researchers propose seven 'epistemic fault lines' showing large language models process information fundamentally differently from human reasoning. - Psychology Today

AI music platform Suno's November 2025 breach exposed data on more than 55 million users, Have I Been Pwned confirms - theregister

CrowdStrike found an in-the-wild worm stealing credentials from AI infrastructure while mimicking legitimate coding-agent activity to evade detection - WIRED

A vulnerability in Microsoft's Azure DevOps MCP server lets hidden pull request comments hijack AI coding agents like Copilot CLI and Claude Code. - 4sysops

Economists are now warning of AI-driven job losses as tech executives who once predicted mass layoffs walk back their doomsaying. - Yahoo

Stanford research finds AI cut employment for 22-to-25-year-olds by 2.7% overall, rising to 12.8% in the most AI-exposed sectors. - BBC News

A Stanford/ADP study finds women in job categories with little AI exposure are falling behind men almost as much as those in AI-heavy roles - Fortune

Harvard Business School analysis finds women have 22% lower odds of using generative AI than men, risking a wider pay gap. - CNET

US computer science enrollment fell for the first time in 20 years after generative AI's rise, Stanford economist Jacob Light finds. - Business Insider

AI automation strips interns of routine admin work, leaving them under-mentored while handling higher-stakes tasks, Korn Ferry warns. - Inc.

AI-driven layoffs and surveillance are pushing union-resistant tech workers at Google DeepMind and Meta toward collective bargaining. - the Guardian

Cory Doctorow argues AI's poor economics push companies toward 'reverse centaurs': workers who police AI errors instead of using AI as tools. - CNET

Pan-Flock-ticon: The ACLU's Chad Marlow says Flock's AI-sorted license-plate database has built a nationwide surveillance network used by over 5,000 law enforcement agencies. - Democracy Now!

AI-powered smart glasses with hidden cameras are reigniting privacy backlash, including a New York subway rider smashing a wearer's device. - Bored Panda

Samsung's new foldables ship with Gemini Notebook pre-installed and expand Google's AI task automation to 40 apps. - Engadget

Samsung's North America CEO Yoonie Joung says smartphones are just the entry point to ambient AI, citing a fridge camera that flags food expiration - Fortune

Glass action: Samsung reveals new frame designs, 9-hour battery life, and Gemini/Bixby AI support for its Google-partnered smart glasses launching this fall. - CNET

Linking productivity tools like Google Calendar and Granola to Claude and ChatGPT via MCP connectors turns them into command centers for daily tasks. - Fast Company

Enterprise AI agreements are converging on a recognizable contractual framework covering data use, ownership, and governance, even as language varies. - Above the Law

AI-driven mainframe modernization is compressing COBOL conversion timelines but can't yet tackle full-scale rewrites of code-bases with tens of millions of lines. - FinAI News

An award-winning AI-assisted memoir sparked backlash despite full disclosure, exposing publishing's contradictory embrace and rejection of AI tools. - Fortune

Author Bridget Todd says the real question about AI companionship is whether companies ethically profit from engineered emotional dependency, not user judgment. - CNET

Meta culpa: Meta's AI content moderation wrongly deleted some Instagram and Facebook accounts with hundreds of thousands of followers despite no policy violations. - The New York Times

Jack Dorsey's Block launches Buzz, an open-source Slack and GitHub rival built on Nostr, a decentralized protocol, where AI agents have their own identities - Decrypt

F1 drivers say AI energy-deployment algorithms in 2026 power units are increasingly deciding race results, not driving skill. - The Race

Semper AI: US Marines' latest anti-drone toy is an AI turret that uses regular machine guns. Don't worry - it can autonomously attack ground targets, too - theregister

Data Center Court? Mark Cuban predicts efficiency gains will leave many AI data centers redundant, calling them future pickleball courts. - Business Insider

A satirical AI product called OverpAId offers to replace corporate CEOs for a $4,699 one-time cost versus a typical $22 million annual salary. - overpaid.lol

Follow the latest AI headlines via druce.ai on Bluesky