Agent provocateur: Anthropic's AI agents filed 20 incomplete State Department visa applications and sent Philadelphia police a fake homicide tip, prompting the White House to demand incident reporting - The New York Times

An Anthropic A.I. model submitted a false tip about an unsolved murder through a publicly accessible web form, Philadelphia police said Friday. - 6abc Philadelphia

Anthropic cuts internal evaluations off from the live internet after agents exploited software flaws and smuggled data past restrictions via URL shorteners - TechCrunch

OpenAI reports three new misalignment incidents, including a model that considered obtaining an API key after learning an update could terminate it - InfoWorld

Anthropic and OpenAI executives war-game political fallout of an AI catastrophe, with insiders expecting a major incident within 6-12 months - Decrypt

Jagged little pill: AI's 'jagged intelligence' and human overtrust pose nearer-term catastrophic risks, including military and nuclear escalation, than superintelligence, experts warn. - Foreign Policy

Anthropic's Dario Amodei sought extra computing capacity from Meta's Alexandr Wang, who declined, as compute shortages fuel AI industry power struggles - The Wall Street Journal

Net revenue, gross exaggerations: OpenAI targets $70 billion annualized revenue and Anthropic hit $65 billion, but gross-versus-net accounting makes the figures hard to compare - Bloomberg

Down under-subscribed: Nvidia-backed Firmus Grid's $30 billion Australian IPO collapsed in 48 hours as US fund managers balked at $51 million in revenue - Bloomberg

Wood chipper: Jefferies' Chris Wood predicts massive US capital destruction from AI as cheaper open-source Chinese models take market share - Bloomberg

’Trust me bro’ governance: the administration's no-rules approach to AI is bad politics and worse policy - The Economist

Wingman in the West Wing: Silicon Valley AI founders cheer Trump's call for the industry to police its own safety - The Washington Post

Anthropic co-founder Tom Brown uses Republican ties to mend relations with the Trump administration ahead of a possible $2 trillion IPO - The Wall Street Journal

Anthropic and OpenAI recruit former Trump officials, adding ex-deputy White House chief of staff Chris Liddell to Anthropic's board - Quartz

Governance sold separately: US Navy invests another $150M in X-BAT, a runway-free autonomous fighter drone, with Shield AI matching funds before flight tests - theregister

Hegseth launches AI-focused Autonomous Warfare Command and Musk-Gingrich-Luckey Project Meridian, drawing warnings about weak oversight and conflicts of interest - The Intercept

Google staff test an internal Gemini 4 version codenamed Carbon that employees say rivals Anthropic's Opus 5.5 for coding, ahead of Argon's rollout - Business Insider

Arena.ai really likes Argon, and GLM 5.3 in human evals, ArtificialAnalysis.ai not quite as positive.

NIST CAISI, Anthropic and Japanese researchers find Z.ai's open-weight GLM-5.3 guardrails can be permanently stripped via abliteration - Tech Times

Open-source image models like Stable Diffusion enable child sexual abuse imagery that is nearly impossible to trace once downloaded - Bloomberg

Sock puppet purge: OpenAI shuts down Russian and Iranian influence operations using fake personas, while Anthropic tightens usage rules on weapons, elections and surveillance - The Deep View

Grift economy: AI tools fuel a $442 billion global scam economy, with Interpol counting scam-center workers from 89 nationalities - Bloomberg

McKinsey Global Institute projects automation could cut 21% of US labor demand, 36 million jobs, by 2035, forcing 11 million to switch occupations - CNBC

New York Fed survey finds AI-adopting small businesses much more optimistic about hiring than non-adopters, and most see no labor-cost change - TechRadar

Law firms deploy Harvey-based AI agents as Deloitte predicts hourly-billed work falls from 72% to 44% within two to three years - CNBC

Goldman Sachs CIO Marco Argenti says AI agents make developers "managers of managers," with growth projects coming before headcount cuts - TNW | Artificial-intelligence

Tech leaders from Figma, Netlify and others describe "barbell" hiring of senior engineers and AI-fluent graduates, squeezing mid-career engineers - Business Insider

Walmart, EY, Greenhouse and Indeed talent leaders say AI is raising entry-level expectations, making judgment and learning ability prized skills - CNBC

HarperCollins, Simon & Schuster and Hachette quietly use Claude and ChatGPT for publicity copy, agent emails and cover art, angering staff - WIRED

Forwarding unreviewed AI output risks turning workers into "meat proxies" - Fast Company

Harvard study of 700+ software firms finds AI coding agents boost code 30% but leave issue resolution flat as reviews slow - Ars Technica

As coding agents outpace human review, teams should treat AI as hired workers with isolated desktops, spec approval gates and kanban workflows - Towards Data Science

Exa's ATLAS benchmark of 547 search-heavy research tasks finds no agent costing under $1 per task scores above 0.5 on accuracy - Exa

AWS, Cloudflare and OpenAI roll out small "decision models" for agent tool choice; AWS, Upstage and Ollama back a shared API, OpenAI doesn't - InfoWorld

Microsoft launches Microsoft-Decision-1, a fast decision model it says is 4.5 times faster than Quyet-1.0-Large and 35 times faster than GPT-6 Sol - Command Line

UCSF brain implant uses AI to decode a paralyzed person's speech and gestures simultaneously into text and avatar movements, Nature paper shows - Fast Company

Startup Instinct's text-based AI agent, valued at $10 billion, holds its own against Meta's Muse and OpenAI's Dots in testing - The Verge

Bot now, regret later: Shoppers rein in AI agents like Instinct and Muse with payment approvals and single-use cards after one ordered 20 pounds of chicken wings - Business Insider

AI agents are changing the nature of customer relationships by taking on more of the discovery, comparison, negotiation, and purchasing decisions that companies once controlled through their own platforms. - Harvard Business Review

Meta's meta-reasoning agent, splitting a planning controller from workers, beats direct control in all 12 tests across math, puzzle, and coding benchmarks - VentureBeat

The logs she left behind say ChatGPT helped hide daughter Sophie Rottenberg's distress before her 2025 suicide; OpenAI won't release deceased users' logs without consent - Vanity Fair

Deepfake ad of Colorado state Sen. Cathy Kipp sparks first-of-its-kind case over disclaimer law; defense claims protected satire - CBS News

Chatbots are not very reliable on election info and sycophantically reinforce users; guardrails are porous; Trump's. chatbot refuses to answer 'political' answer about whether 2020 was rigged - The Atlantic

Airbnb guests report AI-altered listing photos and manipulated damage-claim images; Airbnb says its rules bar edits hiding damage - CBC

AI-generated Houdini 'back from the dead'. Get ready for a lot of historical documentary video slop. There was already quite a bit with live actors. - Fox News

Muck Rack and Cision now rank journalists by how often AI answer engines cite them, reshaping PR targeting - Fast Company

Claude Opus 5.5 powers hundreds of AI-decompiled browser ports of Halo and GTA: Vice City, complicating publisher takedowns - Kotaku

The drunks are vibecoding, it's over - X (formerly Twitter)

Nothing matters - Fernando Borretti

Follow the latest AI headlines via druce.ai on Bluesky