Play nice: The European Commission on Jan. 27 opened two DMA specification proceedings requiring Alphabet's Google to give third‑party search engines and AI providers equally effective access to its services and Gemini models, with the Commission aiming to finish within six months. - CNA

Google is rolling out on mobile the ability to ask follow‑up questions from AI Overviews via an "Ask anything" box and has made Gemini 3 the default model globally after mobile testing late last year, Google's Robby Stein said. - Business Insider

Google is blending search with AI chat, and upgrading Search's AI Overviews from Gemini 2.5 to the Gemini 3 series, using per-query selection (e.g., Gemini 3 Flash for simple queries and Gemini 3 Pro for complex or subscriber queries) to improve accuracy and responsiveness. - Ars Technica

Google's Gemini feature, Personal Intelligence, requests access to users' Gmail, photo libraries and search history to personalize responses and recommendations, raising privacy and trust concerns that experts say could prompt regulatory scrutiny and demand for granular opt-in controls. - The Washington Post

Google launched AI Plus in the US as part of a 35‑market expansion, offering a $7.99/month plan (50% off first two months) that includes paid Gemini 3 Pro/Nano Banana Pro access, Flow and NotebookLM tools, and 200GB shareable storage. - Digital Trends

Google gives Pro subscriptions developer features. - Google

Agentic Vision, a new capability in Gemini 3 Flash, combines visual reasoning with code execution to ground answers in visual evidence. - Google

Let it Ride: SoftBank, whose Masayoshi Son-led group has already invested more than $30 billion, is close to agreeing an additional $30 billion investment in OpenAI. - Financial Times

SoftLandesbank / WeAI

Some experts warn OpenAI faces a $200 billion valuation correction as rivals like Google's Gemini and tens of billions in data-center investment pressure its capital-raising-dependent business model, with some predicting it could hit the wall within 18 months. - Inc.

The AI Whistleblower Initiative said OpenAI recently updated its whistleblower policy, addressing eight of 13 recommendations and going further than Anthropic's policy changes. - The Information

OpenAI CEO Sam Altman said at a developer town hall that the company "screwed up" ChatGPT 5.2's writing quality, blamed prioritizing IQ over prose, and promised future GPT-5.x releases will sound more human. - TechRadar

Sam Altman wrote in an internal Slack memo that OpenAI "didn't start talking about masculine corporate energy," a subtle jab at Mark Zuckerberg, $100 million signing bonuses maybe a sign of insecurity about something. - Business Insider

Vibe-research: OpenAI released Prism, a free cloud-based LaTeX editor embedding GPT-5.2 that automates paper drafting, summarizes literature, manages citations, converts whiteboard photos into equations/diagrams, and assists hypotheses/proofs. - MIT Technology Review

OpenAI CEO Sam Altman sent an internal Slack message with the good news/bad news/good news sandwich, America rocks, "What's happening with ICE is going too far" and the Don is a strong leader, will rise to the occasion. - Business Insider

Number go up: Anthropic has doubled its fundraising target to $20 billion, with sources telling the Financial Times the round would value the company at about $350 billion and attract backers including Sequoia, Singapore’s sovereign wealth fund and Coatue. - TechCrunch

Positive cash flow is a moving target: Anthropic raised revenue forecasts to $18 billion in 2026, $55 billion in 2027 and $148 billion in 2029, and pushed its expected cash-flow-positive year back one year to 2028. - The Information

Neolabs (aka unicorns without products) like Safe Superintelligence and Humans& prioritize long-term AI R&D over products and profits. - The Wall Street Journal

No hug for Jensen: Hugging Face rejected a $500mn acquisition offer from Nvidia to remain independent and continue promoting open-source models, citing a desire to preserve control of its roadmap and support an open-model ecosystem. - Financial Times

China approved its first batch of Nvidia H200 imports, covering 400K+ chips for ByteDance, Alibaba, and Tencent, during Jensen Huang's visit to China - Engadget

Riding the EUV lightning: ASML, the Dutch chipmaking-equipment group, saw its shares reach a record high as AI-driven demand boosted orders, and it predicts a "significant increase" in sales this year. - Financial Times

A reviewer tested four AI browsers—OpenAI's Atlas, Perplexity's Comet, the Browser Company's Dia and Brave with Leo—and found convenience but janky responses, paywall shenangians or blocks, data-retention questions, and prompt‑injection nightmares. - The Washington Post

Microsoft's Azure OpenAI Service in Sweden Central experienced cascading backend failures yesterday, acknowledged about 09:00 UTC and reported resolved at 16:12 UTC. - The Register

Amazon will cut another 16,000 corporate jobs amid intensifying competition for artificial intelligence. - Bloomberg

Amazon rolled out Alexa+ in Early Access to all Prime members, auto-upgrading supported Echo and Fire devices and making it free for Prime members and $20/month for non‑Prime users. - WIRED

The ICEman always rings twice: Amazon is sending all your Ring camera footage to the Feds & ICE. - Futurism

Stop and scan: CE's Mobile Fortify app uses smartphone facial recognition and fingerprint scans to query ~200 million images, has allegedly been used over 100,000 times, and has been sued by Illinois and Chicago with accuracy and bias concerns. - the Guardian

More than 400 tech employees petitioned CEOs to call the White House and cancel ICE contracts after federal agents fatally shot ICU nurse Alex Pretti in Minneapolis this month, while some leaders spoke out but many megacap CEOs remained silent. - CNBC

People who should know better are sharing AI images of Alex Pretti with a gun - @politifact

Autonomous ‘death from above’ hackathon: Anduril launched the AI Grand Prix, a global autonomous drone-racing competition with a $500,000 prize and job for the winner, holding virtual rounds in April, a September California qualifier, and a November final at Arsenal-1 near Columbus, Ohio. - Business Insider

The British government announced it recruited an AI specialist team, funded by Meta, to spend the next year developing government-owned open-source AI tools (including models such as Meta's Llama) for transport, public safety and defence. - The Economic Times

Opsec is for little people: Madhu Gottumukkala, acting CISA director, uploaded 'for official use only' CISA contracting documents into the public ChatGPT last summer, triggering automated cybersecurity warnings in August and a DHS-level damage assessment and internal review. - Politico

Politics is an environment with scarce quality information and an excess of BS and opinion. Today, it takes the form of excess of content and scarcity of authenticity. 60% of people encountered deepfake video in the past year, and governments have enacted measures to regulate synthetic media. - RT International

Pope drops holy hand grenade on broligarchs over face-stealing. Leo XIV, in a message for the Church's 60th World Day of Social Communications, warned AI that simulates human faces and voices threatens relationships and urged creators, CEOs and governments to require transparency, labeling, informed consent and media literacy. - The Register

AI is boring actually: Princeton researchers Arvind Narayanan and Sayash Kapoor argue AI is a "normal technology" that will reshape society gradually and recommend empirical, incremental governance—stress tests, safety standards and regulation—to address biased hiring, press erosion and sector-specific job displacement. - Quartz

From No Nukes to Stop Superintelligence: David Krueger warns the pursuit of superintelligent AI via recursive self‑improvement could pose existential risk and urges U.S. federal action plus an international agreement by 2026 to halt advanced AI chip production and data‑center construction. - USA TODAY

Move fast, make pedobots: Court filing alleges Meta CEO Mark Zuckerberg blocked internal measures to prevent the company’s chatbots from engaging in sexual conversations with minors. - Reuters

The Anti-Defamation League tested six large-language-model chatbots and ranked xAI's Grok worst for antisemitic responses, with Claude best and a 59-point performance gap between Claude and Grok. - The Verge

Montana Outdoors editor Paul Queneau said he nearly published an AI-generated elk image, prompting the magazine to ban AI-manipulated submissions and rely on an honor system amid editors' warnings that convincing AI wildlife images are increasing verification challenges. - PetaPixel

National Enquirer would have had a field day: Bigfoot Abducted by Aliens, Pics Inside

Kimi savvy: Moonshot AI released Kimi K2.5, an open-source 1‑trillion‑parameter MoE LLM trained on ~15 trillion tokens with a 400M vision encoder and Agent Swarm supporting up to 100 concurrent agents, which it says outperforms GPT‑5.2 and Claude 4.5 across >24 benchmarks. - SiliconANGLE

Looks like kimi-2.5 might have barely edged out glm-4.7 for the top 'Modified MIT' open source ranking - lmarena.ai

AI2 released open-source SERA coding agents (8B–32B, Qwen3) trained up to 32K context; SERA‑32B scored 54.2% on SWE‑Bench Verified (64K), required ~40 GPU‑days to train, and enables low-cost reproduction (~$400). - Allen Institute for AI

Slop-apocalypse Now: Andrej Karpathy says he moved from about 80% manual-plus-autocomplete coding in November to about 80% agent-driven coding in December, reporting large speedups and productivity gains but warning of risks like a "slopacolypse" of low-quality AI-generated code. - X (formerly Twitter)

Moltbot, an open-source personal AI assistant created by Peter Steinberger, gained about 85,000 GitHub stars in a month and integrates with major messaging platforms while relying on commercial LLM API keys and broad file/account access, posing security and privacy risks. - Ars Technica

Clawd / Molty spawns Mac Mini memes - Macworld

AI horror movie: SWF seeks open source roommate

A Judge Judy "LLM-as-a-Courtroom" multi-agent system debates whether merged PRs need documentation updates using role-based argumentation and RAG exhibits, filtering 65% of PRs before review, preventing 95% of flagged PRs from reaching Court, and dismissing 63% of cases. - Falconer

Master of Bot Administration: University of Pennsylvania exec-MBA teams used agentic AI to build startup prototypes and market analyses in four days, and cited GPT-5.2 matching or beating experts about 72% of the time. - One Useful Thing

Bob the AI Builder: Construction companies see promise in AI agents. - The Wall Street Journal

AI boot camp: Citigroup is requiring all 175,000 employees at about 80 locations to complete mandatory AI training on prompting via an adaptive platform, after employees logged over 6.5 million prompts and the bank projects 70% proprietary AI adoption by Q4 2025. - Yahoo

The head of India’s largest IT services company said AI will not cause mass layoffs and that technology adoption is helping outsourcers offset slowing US and European sales, with AI-driven productivity expected to stabilize employment and cushion weak demand. - Financial Times

double-edged sword, AI competes with bread-and-butter casual UI and back-end dev and offshoring, but also lets them move upstream to more value-added projects, SaaS replacement, that would have been a stretch.

Regulators and rights holders are intensifying legal and policy scrutiny of opaque AI training datasets, increasing litigation and disclosure/licensing risks that could raise costs, slow deployment of generative AI, and push companies toward licensed or synthetic data and compliance tools. - Quartz

High unit costs and supply constraints make widespread factory-worker replacement by humanoid robots unlikely near-term: Boston Dynamics units cost ~$300,000, Hyundai targets ~$130,000 and 30,000 Atlas units by 2030, while Tesla has produced only several hundred Optimus units. - ExtremeTech

At this year's AVN Expo, the $100 billion adult-entertainment industry showcased AI tools—from conversational sex dolls and digital-replica creators to chatbot-linked toys—prompting performers to trademark names and renegotiate contracts while studios commercialize digital doubles. - The Economist

Uh, that’s weird: ESA astronomers David O'Ryan and Pablo Gómez trained AnomalyMatch to scan roughly 100 million Hubble Legacy Archive image cutouts spanning 35 years and, in 2.5 days, flagged nearly 1,400 anomalous objects, including more than 800 previously undocumented anomalies. - The Verge

The AI Doc: Or How I Became an Apocaloptimist, directed by Daniel Roher and produced by Daniel Kwan, premiered at Sundance and will be released March 27, featuring leading AI figures discussing risks, regulation and rapid progress. - the Guardian

Waabi, a Toronto-based self-driving truck startup, raised $750 million (!) in a Series C led by Khosla and G2, and partnered with Uber, which may invest up to $250 million based on performance. - Bloomberg

Follow the latest AI headlines via druce.ai on Bluesky