AI News Roundup: January 12-19, 2026

The week of January 12-19, 2026 marked escalating healthcare AI competition with OpenAI’s $60-100M acquisition of Torch Health, Anthropic’s Claude Healthcare launch, and Google’s Personal Intelligence beta. Meanwhile, AI models achieved breakthrough progress solving 15+ Erdős mathematics problems, Anthropic revealed AI will worsen global inequality, and the startup raised a reported $25 billion at $350 billion valuation. ChatGPT introduced ads to free tiers, xAI raised $20 billion, and Andrea Vallone departed OpenAI’s safety team for Anthropic—the latest in a string of high-profile safety researcher defections. FTC reversed its first AI enforcement action under the new administration’s AI Action Plan.

🧠 Big Tech News

OpenAI acquires Torch Health for $60-100M to power ChatGPT Health

OpenAI acquired one-year-old healthcare startup Torch Health on January 12, with reports varying from $60 million (CNBC) to $100 million (The Information)—paid primarily in equity. Torch built “unified medical memory” aggregating fragmented health records from hospitals, labs, wearables, and consumer testing. The team joins OpenAI to support ChatGPT Health.

Anthropic launches Claude for Healthcare on January 12

One day before OpenAI’s Torch announcement, Anthropic launched Claude for Healthcare—a HIPAA-ready product suite enabling providers, payers, and consumers to use Claude for medical purposes. The move intensifies competition as AI companies cluster around the 40 million daily healthcare queries OpenAI reported.

Anthropic launches Claude Cowork for non-technical workflows

Anthropic released Claude Cowork on January 13 as research preview for macOS Claude Max subscribers, extending agentic capabilities beyond coding to everyday tasks. Cowork can read, edit, and create files across local folders—reorganizing downloads, building spreadsheets from screenshots, and drafting reports—with user-controlled permissions.

Google introduces Personal Intelligence beta for Gemini

On January 14, Google launched Personal Intelligence beta allowing Gemini to connect with personal Gmail, Photos, Search, and YouTube history data with explicit user opt-in. The feature makes responses more tailored and context-aware, initially available to U.S. users.

OpenAI announces ads in ChatGPT Free and Go tiers

OpenAI revealed on January 15 that it will introduce clearly labeled ads in ChatGPT Free and Go tiers while keeping Pro, Business, and Enterprise ad-free. The company emphasized user privacy and stated ads won’t influence responses—marking a notable shift toward sustainable monetization.

⚖️ Politics & Legal Affairs

White House science chief calls state AI laws “anti-innovation”

In January 14 House Science Committee testimony, OSTP Director Michael Kratsios urged adoption of a single national AI rulebook, calling patchwork state laws “anti-innovation.” Democrats pressed him on budget cuts to science agencies and the administration’s push to limit state-level AI rules.

Congress debates AI Action Plan implementation and regulation

House subcommittee hearing on January 14 pressed the White House for answers on AI Action Plan implementation amid safety concerns and backlash over tech companies’ expanding footprint for AI infrastructure. Lawmakers debated where Congress needs to step in.

UK to criminalize non-consensual intimate AI images

The UK will bring into force a law making it illegal to create non-consensual intimate images, following widespread concerns over Elon Musk’s Grok AI chatbot generating such content. The law represents first-of-its-kind regulation targeting AI-generated explicit imagery.

🔬 Research & Development

AI models solve 15+ Erdős mathematics problems in breakthrough

Since Christmas, 15 problems moved from “open” to “solved“ on the Erdős website, with 11 solutions crediting AI models. Mathematician Terence Tao documented eight cases where AI made meaningful autonomous progress, with GPT-5.2 solving problems after 15-minute reasoning sessions verified with formal proof tools.

LMArena raises $150M Series A at $1.7B valuation

LMArena, which started as UC Berkeley research project in 2023, announced on January 13 a $150 million Series A at $1.7 billion post-money valuation, led by Felicis and UC Investments. The platform has become the industry-standard model evaluation benchmark.

🌍 Tools and Launches

Anthropic Claude Cowork introduces agentic file management

Claude Cowork enables users to grant Claude access to chosen folders for reading, editing, and creating files. Users can queue tasks like reorganizing downloads, building spreadsheets from screenshots, or drafting reports, with explicit folder and connector permissions.

Infosys generates 28 million lines of code using AI tools

CEO Salil Parekh announced January 14 that Infosys has generated more than 28 million lines of code using AI tools, with company attrition dropping to 12.3% compared to TCS’s 13.5% and HCLTech’s 12.4%.

🌍 Startups & Investments

Anthropic reportedly raising $25B at $350B valuation

Reports emerged January 18 that Anthropic is advancing new financing round to raise at least $25 billion at $350 billion valuation, with Sequoia Capital planning to participate alongside existing investors Microsoft and NVIDIA. Transaction expected within weeks, representing >90% valuation increase in four months.

xAI completes upsized $20B Series E funding round

Elon Musk’s xAI completed Series E funding exceeding the $15 billion target, raising $20 billion total. The round positions xAI to compete directly with OpenAI, Google DeepMind, and Anthropic in frontier AI development.

osapiens reaches unicorn status with $100M Series C

Sustainable growth enterprise software provider osapiens secured $100 million Series C financing on January 15, pushing the company to unicorn status ($1B+ valuation). The round reflects growing investor interest in AI for ESG and sustainability applications.

🌍 AI News in EU & Sweden

Meta pauses Ray-Ban Meta AI glasses in four European countries

Meta’s decision to pause launches in France, Italy, Canada, and UK reflects ongoing regulatory and supply chain challenges in European markets. The pause comes amid continued EU scrutiny of Meta’s AI practices.

🧠 AI in Healthcare & Education

ChatGPT Health developed with 260+ physicians over two years

OpenAI revealed ChatGPT Health was developed in close collaboration with more than 260 physicians worldwide over two years to provide clear and useful health information, demonstrating commitment to clinical validation.

🤖 Robotics

Boston Dynamics Atlas humanoid continues production development

Updates indicate Boston Dynamics’ electric Atlas humanoid robot continues production development with Hyundai partnership and Google AI research lab collaboration (not OpenAI or Anthropic). Robot demonstrated at CES continues testing phase.

🎮 Hardware

Anthropic commits to $30B Azure computing capacity with NVIDIA chips

As part of November 2025 deal continuing through January, Anthropic committed to purchasing $30 billion worth of Azure computing capacity featuring NVIDIA Grace Blackwell and upcoming Vera Rubin architecture.

📊 Market Insights & Investment Trends

NVIDIA becomes “hidden winner” in OpenAI-AWS and Anthropic-Azure deals

Analysis shows NVIDIA positioned as key infrastructure layer in AI buildouts, with OpenAI’s $38B AWS deal and Anthropic’s $30B Azure agreement both featuring NVIDIA chips. Company set for continued revenue acceleration through 2026 and beyond.

🧠 Adoption Trends & Consumer Behavior

84% of enterprises expect AI low-code/no-code platforms to scale

Survey found 84% of enterprises expect AI-powered low-code and no-code platforms to scale inside their businesses within next 18 months, representing shift toward democratized development tools.

🧠 Research Paper of the Month

On the Slow Death of Scaling

Sarah Hooker Available on https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5877662

The essay argues that “bigger is better” has dominated AI for a decade but is an increasingly fragile, distorted guide for real progress in the field.​

Core argument

• The author challenges the idea that simply scaling model size and training compute will continue to drive the frontier, arguing that the performance–compute relationship is more uncertain and rapidly changing than current hype suggests.​
• Dependence on scaling has reshaped incentives, concentrating power and capital in a few industry labs and marginalizing academia and smaller actors from frontier AI research.​

Historical context

• The essay uses analogies like the magnetron → microwave story to show how repurposed technologies can suddenly unlock new domains, comparing this to GPUs being repurposed from graphics to deep‑learning accelerators.​
• GPUs’ ability to massively parallelize matrix multiplications (higher FLOP/s and distributed training) unlocked deeper neural networks and led to dramatic performance jumps on benchmarks like ImageNet between 2011–2015.​

Scaling and the “bitter lesson”

• The piece traces how a sequence of results (e.g., AlexNet and subsequent deep CNNs, large‑scale LMs) triggered a “bigger‑is‑better” race in parameters, datasets, and training cost, pushing compute demands up by orders of magnitude.​
• It engages with Rich Sutton’s “bitter lesson”: brute‑force compute and general learning methods have historically beaten hand‑crafted, symbolic domain knowledge—while questioning whether extrapolating that logic into the future is still justified.​

Consequences for the research ecosystem

• Charts from the AI Index report are used to show steep growth in training costs for frontier models (2016–2023) and the concentration of “notable machine learning models” in a few geographies, reflecting capital and compute centralization.​
• The essay argues that this concentration and cost explosion have distorted research culture: industry labs publish less, academic participation at the frontier has dropped, and many alternative innovation paths are under‑explored.​

Forward‑looking claim

The author concludes that future breakthroughs are unlikely to come from scaling alone and that new levers of progress—beyond raw training compute—are emerging, implying that significant disruptions to the current scaling‑centric paradigm lie ahead.​

🧠 Tools to Try

Claude Cowork (Anthropic)

Agentic file management tool extending Claude Code capabilities to non-developers. Grant Claude access to chosen folders for reading, editing, and creating files. Can reorganize downloads, build spreadsheets from screenshots, draft reports from notes. Available as research preview for Claude Max subscribers on macOS.

ChatGPT Health (OpenAI)

Upload medical records from Apple Health, Function, MyFitnessPal, and other wellness apps for personalized health guidance. Developed with 260+ physicians over two years. Designed to support, not replace, medical care—helps interpret test results and create exercise plans.

Claude for Healthcare (Anthropic)

HIPAA-ready product suite for healthcare providers, payers, and consumers. Enables medical purpose usage with regulatory compliance. Targets healthcare startups and large enterprises integrating AI into clinical operations. URL:

Gemini Personal Intelligence (Google – Beta)

Connect Gemini to personal Gmail, Photos, Search, and YouTube history with explicit opt-in. Provides tailored, context-aware responses based on your data across Google services. Currently available to U.S. users in beta.