Too Fabulous to Function: Fable 5 crossed a line the world was not ready for. - Fast Company

Bipartisan House members demanded answers from the Trump administration on why Anthropic's Fable models were restricted, and whether rivals face similar treatment - The Washington Post

The White House and Anthropic are drafting shared benchmarks to measure how dangerous AI jailbreaks are, as export controls remain in force - Politico

Glasswing and a prayer: Dragos, Cisco, and roughly 200 other organizations in Anthropic's Project Glasswing program retained Mythos access; European agency ENISA was cut off - TNW | Anthropic

Amazon MGM Studios drops Luca Guadagnino's mostly finished movie starring Andrew Garfield as Sam Altman; Amazon struck a major deal with OpenAI in February, including a $50B investment - Variety

Odd lots of doom: Anthropic co-founder Jack Clark and head economist Peter McCrory discussed recursive self-improvement preparations and existential AI risk on the Odd Lots podcast - Bloomberg

NewEdge Advisors deploys Anthropic Claude to 450-plus advisors for portfolio analysis and meeting prep. - Wealth Management

GPT 5.6 rumored next week - TestingCatalog AI News

Barret Zoph exits OpenAI again, just five months after rejoining to lead enterprise AI sales - The Verge

AI's $660B infrastructure boom runs into a power grid that takes a decade to expand - Yahoo

FERC approved orders to cut power grid connection wait times for AI data centers from years to 90 days. - Bloomberg

Goldman Sachs warns a "Postmodern Cycle" of surging capex has replaced buybacks, with hyperscalers set to spend $755 billion in 2026 - Yahoo

Buybacks are so 2022: Big Tech buybacks collapse as AI spending crowds out shareholder returns, with only Microsoft repurchasing shares in Q1 2026 - Bloomberg

Posphide and seek: China's export controls on indium phosphide — a key material for AI data center optical links — have driven prices up 250% since 2025 - TNW | China

Amazon is in talks to sell its custom AI chips to other companies' data centers to compete with Nvidia. - Bloomberg

Over 100 rights groups including Amnesty International warn that military AI systems in Gaza enable civilian casualties and evade accountability. - NaturalNews.com

Google released an AI Control Roadmap treating its own AI agents as potential insider threats, with human supervisors monitoring agent actions in real time - Google DeepMind

Socket to me: Malicious webpages can open WebSocket connections to local MCP servers in Microsoft AutoGen dev builds, enabling full host-level code execution - 4sysops

RCE la vie: Four AI security incidents in one week: Copilot mailbox exfiltration, LiteLLM admin takeover, Langflow RCE on 7,000 instances, and an npm worm - VentureBeat

US government and companies race to lock down data as AI creates new attack surfaces and compliance demands - Bloomberg

Google backed a $3.2 billion New York data center renting its TPUs to Anthropic, mirroring Nvidia's strategy for stimulating AI chip demand - The Wall Street Journal

SpaceX shares plunge 20%, erasing $600 billion in market cap after its $60 billion all-stock acquisition of AI coding startup Cursor - Forbes

Mars but no margins: SpaceX's $2.4 trillion valuation is driven by far-off plans including space data centers and a Mars settlement. - The Washington Post

Yann LeCun calls xAI "a failure" and warns of a bubble explosion as top AI labs burn cash - CNBC

EUV gotta be kidding: Commerce Secretary Lutnick warned ASML that China may have illegally obtained an EUV chip-making machine, in its most significant confrontation with Washington over export compliance - Bloomberg

America Is Headed Toward the Infinite Workweek - The Atlantic

Use it or lose it: AI deskilling is real: Polish doctors lost diagnostic accuracy on AI-free days, and early data on software engineers looks similar - Nature

Two-thirds of office workers secretly use banned AI tools, and many have uploaded confidential data to public AI systems - TechRadar

Singapore music studios report project budget cuts of up to 60% as clients shift to AI-generated music and voiceovers. - CNA

Enterprise AI's gold rush is over; governance gaps, not model limits, are killing most initiatives - TechRadar

AI accelerates M&A integration failures — inconsistent data and fragmented governance make AI outputs look authoritative but mislead - TechRadar

AI vendors are abandoning flat subscriptions for metered pricing, meaning heavy users of email drafting, slide-building, and code tools will pay more - Bloomberg

A market research firm warns that unusually low forward price-to-earnings ratios for Nvidia, Broadcom, and Micron signal investor skepticism, not cheap valuations - Business Insider

Harvey's monthly AI token usage jumped 12x to 13 trillion; CEO says companies must now prove ROI on every token - Business Insider

AI tools deployed at BNY Mellon and Morgan Stanley flag toxic managers in real time, but the same surveillance monitors all employees - Bloomberg

Google AI Overviews presents SCP horror fiction entities as real in at least 20 confirmed cases. - Futurism

Google's new Personal Intelligence system builds behavioral profiles from private Gmail, Photos, and Calendar data to shape search results before you type - Search Engine Journal

MCP's new Enterprise-Managed Authorization extension lets organizations centrally manage tool access through their identity provider, eliminating per-app OAuth prompts - Model Context Protocol Blog

Therap-AI: 35% of psychologists report patients using AI as a secondary therapist, with 15% observing AI-linked delusional thinking - Scientific American

AI chatbots outperformed professional fundraisers in controlled tests, raising concerns about AI-powered persuasion at scale - The Washington Post

NYT Hard Fork podcast debate features Sayash Kapoor and Daniel Kokotajlo on whether AI will transform society gradually or accelerate rapidly. - The New York Times

New Arbor framework outperforms Claude Code and Codex by 2.5× using branching hypothesis trees to preserve context across runs - VentureBeat

strong coding ranking for GLM 5.2 in arena.ai, topping everything but Fable, pretty mid for chat and currently not even better than 5.1. Ofc way cheaper and also strong on agent benchmarks.

TextGrad uses LLM-generated feedback as gradients to auto-optimize AI prompts and code, improving scores on doctoral-level reasoning tests - Hacker Noon

A 2025 University of Pennsylvania study found major inconsistencies across seven AI content moderation systems in detecting hate speech. - Al Jazeera

Ramp-age: Waymo issues its sixth recall, pulling nearly 3,900 robotaxis after software missed freeway ramp-closure signs - Business Insider

“You would not believe the texts I got from these tech guys.” New book: Trump mocked Zuckerberg and Bezos by showing associates their fawning texts, saying they were “kissing my ass” - WIRED

Utah's AI prescription-refill pilot from startup Doctronic drew opposition from the state medical board, which called for suspension on safety grounds. - The Wall Street Journal

Runway's AI film festival wows Lincoln Center crowds, even as critics call the films visually slick but narratively thin - Business Insider

Adobe rolls out AI assistants to Photoshop, Premiere, Illustrator, InDesign, and Frame.io in public beta - The Verge

Venture capitalist Chi-Hua Chien argues AI model commoditization is underway and application companies will capture most value. - Yahoo

Higgsfield launches Supercomputer 2.0 marketing automation agent, claims 78% of Fortune 500 as clients. - TNW | Apps

UPS built a digital twin of its entire global logistics network, updating every 10 minutes, cutting misloads by 70% and labor hours by 9.9% - TNW | Artificial-Intelligence

Follow the latest AI headlines via druce.ai on Bluesky