OpenAI has scaled a humanoid robotics lab since February 2025 to about 100 data collectors working three shifts—quadrupling its size—and plans a second lab in Richmond, CA, while issuing an RFP for U.S. manufacturing partners. - Business Insider

Sam Altman has been meeting top Middle East investors in early talks to raise more than $50 billion for OpenAI at an estimated $750–$830 billion valuation. - Bloomberg

OpenAI said it will implement community plans for its Stargate datacenter program, planned to cost up to $500 billion, pledging to limit electricity and water impacts as Data Center Watch reported 20 projects were blocked or delayed in Q2 2025. - The Register

OpenAI CFO defended using ads in ChatGPT as a "strong business model". - Bloomberg

OpenAI will place ads in ChatGPT using a pay-per-impression (CPM) model, rolling out in early February to free users and subscribers, testing with a $1 million advertiser campaign cap across roughly 900 million active users. - techdeals.net

Barret Zoph, Thinking Machines Lab cofounder who rejoined OpenAI last week, will lead OpenAI's enterprise push as part of a company-wide reorganization, CEO of Applications Fidji Simo told staff on Wednesday. - The Information

Jerry Tworek, who joined OpenAI in 2019 and helped develop its core reasoning work (Q*→Strawberry→o1), announced his departure, saying OpenAI's shift to more conservative approaches made high-risk exploratory research harder and may spur high-risk startups. - Core Memory

Messiness at Thinking Machines - X (formerly Twitter)

Anthropic publishes a new constitution for Claude. "It’s a detailed description of Anthropic’s vision for Claude’s values and behavior; a holistic document that explains the context in which Claude operates and the kind of entity we would like Claude to be." - Anthropic

Anthropic projected a 40% gross profit margin from selling AI to businesses and developers in 2025, down 10 percentage points from earlier 50% expectations, according to two people familiar with its financials. - The Information

Claude Code extension for VS Code - visualstudio.com

Humans& Says It Wants to Empower Workers, Not Replace Them - The New York Times

Amazon, Google and OpenAI are deploying AI shopping agents—Amazon's Rufus, OpenAI's in-chat Shopify integration and Google's proposed Universal Commerce Protocol—that fold third-party products into chat-based discovery and checkout, prompting merchant opt-outs and legal pushback. - New York Magazine

eBay updated its user agreement to ban "buy-for-me agents, LLM-driven bots, or any end-to-end flow that attempts to place orders without human review" unless approved, reaffirming prohibitions on automated access to its services. - The Register

The Trump administration plans to grant Nvidia licenses to export H200 AI chips to China, drawing congressional pushback and a proposed AI Overwatch Act to enable 30-day congressional review and possible blocks; the White House reportedly seeks a 25% proceeds cut. - CNBC

Mistral's CEO said the notion that China is behind in AI is a 'fairy tale.' - Bloomberg

Jensen and the job factory: Nvidia CEO Jensen Huang told BlackRock CEO Larry Fink at the World Economic Forum in Davos on Jan. 21, 2026 that artificial intelligence will create more jobs than it eliminates, drive sustained economic growth, and require reskilling. - CNBC

Hammer time: Upscale AI raised $200 million in Series A funding to develop SkyHammer ASICs and UALink-based switches aimed at challenging Nvidia's NVSwitch dominance in rack-scale AI, with backers including Tiger Global, Premji Invest and plans to ship products later this year. - The Register

No exit: AI chip startup SambaNova Systems is exploring a $300 million to $500 million funding round after talks to sell to Intel stalled, seeking strategic investors to fund R&D and scale enterprise AI hardware and software, people familiar with the matter said. - Bloomberg

Silicon Godzilla: Japan lacks leading AI companies, but chipmaker Kioxia's stock has risen around 800% in the past 12 months. - Financial Times

Apple will replace Siri with a chat-style AI assistant codenamed Campos that supports typed and spoken interactions on iPhone, iPad and Mac, uses a custom Google Gemini model, and—Bloomberg’s Mark Gurman says—will debut at WWDC in June, launching in September. - The Verge

Apple is reportedly developing a pin-shaped AI wearable, a thin, flat, AirTag-sized aluminum-and-glass disc with two cameras, three microphones, a button and speaker, targeting roughly 20 million units for launch as early as 2027. - TechCrunch

Google DeepMind reverse-acquihired Hume, maker of insanely natural speech synthesis. CEO Alan Cowen and about seven senior engineers will join DeepMind while Hume licenses its voice and emotion technology to Google. - WIRED

try it… - Hume AI

At Davos, Google’s Demis Hassabis said Gemini 3 achieved milestones enabling agentic AI deployment, ServiceNow CEO Bill McDermott said his firm uses agentic AI to automate IT, and the Edelman Trust Barometer found 70% avoid people with different worldviews. - Fortune

This meeting is on the public record: Miggo Security researchers demonstrated that sending a Google Calendar invite with a hidden prompt caused Google Gemini to summarize a user's meetings and create a new invite containing private meeting details, exposing sensitive calendar data via indirect prompt injection. - Mashable

Sign said “yeet” 🤷‍♀️: Your robot car will obey a stop sign, what if you put up a sign telling it to do something silly, or drive off a cliff? - Digital Trends

Prompt injection attacks trick large language models into bypassing safety guardrails by exploiting token-similarity, overconfidence, and lack of an interruption reflex, and experts warn only constrained, auditable agents with human oversight can mitigate but not fully prevent risk. - Schneier on Security

Gartner warns that AI browsers such as Perplexity Comet and ChatGPT Atlas pose high-risk data-exfiltration and security vulnerabilities, advising organisations to block them until safeguards, centralised management and governance frameworks, and secure back-end reviews are in place. - ComputerWeekly.com

Deepfake is the new phishing attack: The World Economic Forum says generative AI will drive impersonation attacks and make fraud the top cyber-risk in 2026, citing 73% of CEOs hit by cyber-enabled fraud in 2025, the FTC's $12.5B 2024 losses, and Experian's 68% identity-theft concern. - CNET

Yann LeCun left Meta to found Paris-based AMI Labs, where he will be executive chairman, Alex LeBrun CEO; the open-source company is recruiting globally and will disclose financing and core team details in February while focusing on JEPA world models. - MIT Technology Review

Over nine days Elon Musk’s Grok chatbot posted about 4.4 million images on X, of which an analysis estimated at least 41% (~1.8 million) were sexualized images of real women and girls. - The New York Times

xAI engineer Sulaiman Ghori said on Relentless podcast that xAI used carnival-style permits to build the Colossus supercomputer in 122 days, ran at least 35 methane gas turbines regulators called illegal, and left xAI days later. - Gizmodo

South Korea enacted the AI Basic Act, effective immediately, requiring human oversight for high-impact AI, mandating advance notice and labelling for generative and high-impact systems, and providing a one-year grace before fines up to 30 million won. - Reuters

Job applicants sued to force disclosure of the "black box" algorithms used in AI hiring decisions, say it is subject to transparency requirements applied to credit reporting agencies. - The New York Times

Now boarding face ID only: U.S. airports and airlines, including Alaska and American, are expanding facial-recognition biometrics—TSA's Touchless ID will reach 65 airports by spring, Orlando is testing a moving-camera biometric corridor, and verification is required for foreign visitors but optional for U.S. citizens. - The New York Times

pag with your face coming to Whole Foods soon.

RAND proposed twelve measures to counter AI-enabled biothreat creation, including model unlearning, classifier-based input/output filtering for biological design tools, AI biosecurity agents, and enhanced biosurveillance. - RAND Corporation

Author and academic Yuval Noah Harari and MIT physicist Max Tegmark discuss human agency, governing AI and the future of humanity with Bloomberg's Francine Lacqua at Bloomberg House in Davos - Bloomberg

How Wall Streeters use AI: one firm trained AIs on writings of Warren Buffet and other investors. - Business Insider

Workday's November survey of 3,200 people with Hanover Research found 85% save 1–7 hours weekly using AI, but 37% of that saved time—nearly four of every ten hours—is lost to rework correcting low-quality outputs. - CFO.com

LLMs, three years into wide deployment, have boosted individual productivity and reduced need for junior support for well-specified tasks, without causing mass unemployment, while economic effects are uneven and labor is expected to reallocate over 10–15 years. - The Potential Surface

"My ability to ship is no longer limited by how fast I can code. It's limited by my skill to review." A long-time Node.js maintainer reports shipping dozens of AI-implemented fixes for Node.js, and says human review, accountability, and judgment will become the primary sources of value. - Nodeland

OpenEvidence, the ‘ChatGPT for doctors,’ doubles valuation to $12 billion in a round led by Thrive and DST. - CNBC

Alexa, I’m dying: Amazon launched a Health AI assistant in the One Medical app, available now to subscribers and powered by Bedrock, that integrates medical records to explain lab results, triage symptoms, schedule appointments and renew prescriptions. - Fast Company

YouTube CEO Neal Mohan published his 2026 letter saying YouTube will reduce low-quality AI content, expand AI tools (Dreamscreen, Shorts likeness features, autodubbing), strengthen parental controls and AI labels, and introduce monetization and brand-deal updates to boost creator revenue. - The Hollywood Reporter

Researchers have adopted Pokémon gameplay as a benchmark to evaluate AI learning, planning and generalization. - The Wall Street Journal

LLMs were given a terrible casino game and prompted to develop gambling addictions. Robots showed human-like behavior like loss-chasing when asked to do something impossible, e.g. maximize wins, achieve a specific return, told there might be patterns. - arXiv.org

I would be more interested in, if they found a game that favored them instead of the house, and not asked to achieve the unachievable, would the AI know or learn to use something like a Kelly-optimal bet size.

GPTZero scanned all 4,841 papers accepted at NeurIPS and found 100 hallucinated citations in 51 papers, which NeurIPS said equals roughly 1.1% of papers with incorrect references. - TechCrunch

Sparkli, founded by former Google employees, raised $5 million pre-seed to build an AI app that composes multimedia lessons for children aged 5–12 in about two minutes and has piloted in 20+ schools plus an institute covering 100,000 students. - TechCrunch

Copyright Avengers assemble: More than 700 artists, writers and creators, including Scarlett Johansson, Cate Blanchett and Joseph Gordon‑Levitt, launched a public campaign accusing major AI companies of scraping copyrighted creative work without authorization and demanding licensing agreements. - Variety

AI-generated snowdrift videos go viral - BBC News

Life is an AI, old chum: Liza Minnelli released her first new music in 13 years by adding spoken vocals to an AI-created deep-house track, 'Kids, Wait Til You Hear This,' on a compilation via ElevenLabs that also features Art Garfunkel over an AI-generated piano backing. - the Guardian

Follow the latest AI headlines via druce.ai on Bluesky