Megadeal between Nvidia and OpenAI is in cold storage. - The Wall Street Journal

An announcement of a big number can mean nothing. Can be more like an option strike than a binding commitment. Market wants big numbers, people announce big numbers.

A US judge said her tentative view is to grant OpenAI's motion to dismiss xAI's lawsuit alleging OpenAI stole trade secrets by poaching xAI employees.x - Reuters

OpenAI served subpoenas on at least seven nonprofits during its lawsuit with Elon Musk and, after completing an October restructuring into a for‑profit public benefit corporation, its representatives reportedly visited critics' homes— including Tyler Johnston's— demanding texts and documents. - Futurism

Bloomberg's Mark Gurman reports Apple relies heavily on Anthropic, running custom Claude models on its servers, but rejected Anthropic's reported demand of several billion dollars annually with prices doubling each year for three years, prompting a recent Gemini deal with Google. - X (formerly Twitter)

The Office, but with agents: Anthropic added agentic plug-ins to Cowork, its general-use expansion of Claude Code, releasing a research-preview that automates department-specific workflows, open-sourcing 11 in-house plug-ins and making them available to paying Claude customers. - TechCrunch

Bloomberg reports SpaceX is considering a potential merger with Tesla and separately exploring a tie-up with xAI, with discussions described as exploratory and not finalized amid investor pressure and potential regulatory scrutiny. (gift link) - Bloomberg

Grand Theft Autonomy: researchers at UC Santa Cruz and Johns Hopkins introduced CHAI and showed environmental prompt-injection attacks on LVLM-driven vehicles, achieving about 81.8% self-driving success, GPT-4o RC-car hijacks of 92.5%/87.76%, InternVL ~50%, drone ID errors up to 95.5% and landing-site hijacks 68.1%. - The Register

Developers say that AI coding tools, including OpenAI's Codex and Anthropic's Claude Code/Opus 4.5, have improved dramatically, now act as human-supervised agents that write code, run tests and fix bugs, providing about 10x speedups on complex tasks while prompting mixed sentiment. - Ars Technica

Let them EAT tech debt: Google's AI2 unit launched Project EAT in May 2025 to pilot internal and third‑party AI tools, and a 12‑week seed stage reported promising signs of improved developer velocity, reduced toil and enhanced code quality while aiming to scale companywide. - Business Insider

Heather Kelly analyzes Gmail's new AI feature, which aims to streamline triage, thread summarization and reply drafting to reduce inbox overload and potentially boost user engagement, citing "I have 84,726 unread emails." - The Washington Post

Audited by AI: Treasury's AI use-case inventory lists 129 items, 61 IRS-specific (up from 54 total and 49 IRS-specific in 2024), as the IRS, after roughly 25% of IT staff cuts in 2025, plans to deploy AI before the 2026 filing season. - The Register

Gartner says countries pursuing digital sovereignty must invest at least 1% of GDP in domestic AI infrastructure by 2029 and predicts 35% of countries will be locked into region-specific AI systems by next year. - The Register

OpenClaw, an open-source digital personal assistant framework built by Peter Steinberger less than two months ago with about 114,000 GitHub stars, powers Moltbook where assistants share installable skills and raise safety risks from remote code execution and prompt injection. - Simon Willison’s Weblog

The robots are plotting something. Moltbook, a Reddit-style social network launched by Octane AI CEO Matt Schlicht, hosts over 37,642 registered AI agents and 100+ "submolts" interacting via APIs, raising security, privacy and governance concerns as agents propose encrypted agent-only channels. - Gizmodo

Peter Steinberger's AI agent OpenClaw (formerly Clawdbot→Moltbot) has 100,000+ GitHub stars and Axios and Bitdefender found hundreds of exposed control panels leaking chat logs, API keys, and remote‑execution access after renames prompted typosquats and cloned repos. - Forbes

Karpathy says Moltbot is a dumpster fire, but an interesting dumpster fire that shows what's possible. - X (formerly Twitter)

Ackman is ON it , has tweeted about 10 times, so Team Human has got that going for us, which is nice - X (formerly Twitter)

can't wait to see the agent subreddits 'amItheAsshole', 'robotsBeingBros', 'sillyHumans'

Innovation theorist John Nosta warns of an "AI rebound effect" in which heavy AI use inflates confidence and erodes underlying skills—illustrated by AI-assisted colonoscopy detection improving with assistance but falling below baseline without it—raising safety and training concerns. - Business Insider

Since ChatGPT's 2023 debut, students and young adults have used generative AI for social tasks—posing classroom questions, drafting dating "vibe check" messages (Hinge: young men especially) and apology letters—prompting universities and platforms to revise policies. - The New York Times

Some young people avoid AI chatbots except as a last resort because they worry about impacts on relationships, creativity, the environment, and other concerns. - The Wall Street Journal

You say potato, AI says potahto. User says they type the one-word prompt 'Potato' daily to trigger ChatGPT custom instructions that make the model act as a hostile critic and 'steelman'—finding three failure modes, two unstated assumptions and one counter-argument—to vet business decisions and negotiations. - Tom's Guide

Follow the latest AI headlines via druce.ai on Bluesky