PLUS: Humanoid robots in labs, Google's dangerous AI summaries, and the new class of AI billionaires
Good morning
Anthropic is bringing its AI assistant, Claude, directly onto your computer. A new agent-like feature allows the model to read, edit, and create files, moving beyond simple chat to automate complex tasks right from your desktop.
This marks a significant step toward practical, on-device AI agents that act as active digital assistants. As these tools gain more agency over our personal files, the key question becomes how to balance these powerful new capabilities with critical safety measures.
In today’s Next in AI:
Claude's new agent works on your computer
Humanoid robots enter sterile labs
Google removes dangerous AI summaries
The new class of AI billionaires
Claude's New Coworker
Next in AI: Anthropic is letting Claude work directly on your computer with its new 'Cowork' feature, available now in a research preview on its macOS app. The new agent-like tool enables Claude to read, edit, and create files in a designated folder, letting you automate complex tasks with simple instructions.
Decoded:
Unlike a standard chatbot, Cowork operates with more agency, creating and executing plans while keeping you in the loop—all built on the same foundations as Claude Code but for non-technical tasks.
You can expand Cowork's capabilities by pairing it with existing connectors for external information and new skills that improve its ability to create documents and presentations.
Anthropic highlights that agent safety is an active area of research; while you control folder access, be mindful of potential prompt injection risks and that Claude can misinterpret destructive commands.
Why It Matters: This marks a significant step toward practical, on-device AI agents that go beyond conversation to become active digital assistants. The success of this research preview could shape how professionals use AI for everyday productivity.
The Robotic Clean Room
Next in AI: Startup Multiply Labs is deploying AI-powered robots and humanoids into sterile labs to automate cell therapy manufacturing. This new approach promises to drastically cut costs and scale production for life-saving treatments.
Decoded:
The automation creates a massive cost reduction of over 70% per dose, dropping from over $100,000 to as low as $25,000.
Instead of manual programming, robots learn complex tasks through imitation learning, watching videos of top human scientists to perfectly replicate their expert techniques within NVIDIA's simulation platforms.
To prevent human contamination during loading and unloading, humanoid robots powered by NVIDIA's Isaac GR00T foundation model are being developed to handle materials outside the main robotic work cell.
Why It Matters: This shift moves cell therapy from a complex, artisanal process to an efficient, industrial-scale operation. It also provides a powerful blueprint for how AI-driven robotics can enter other highly precise and sterile manufacturing sectors.
Google's AI Retreat
Next in AI: Google is removing some of its AI Overviews for health-related searches after an investigation revealed they were serving up dangerous and misleading medical advice.
Decoded:
Experts flagged a specific AI summary for liver function tests as 'dangerous' and 'alarming,' as it provided inaccurate results that could lead seriously ill patients to believe they were healthy.
The move highlights the risks of deploying generative AI at scale, especially since Google dominates 91% of the global search engine market.
Despite the removals, health advocates worry the fix isn't comprehensive, as slight query variations and other problematic AI answers on topics like cancer still generate misleading summaries.
Why It Matters: This incident is a critical stress test for AI's role in disseminating high-stakes information. It forces a conversation about the necessary guardrails and the true readiness of generative AI for life-impacting applications.
The AI Aristocracy
Next in AI: An Axios analysis reveals the AI boom is creating a new, rarified class of “Have-Lots” who are accumulating vast fortunes at an unprecedented rate, primarily through exclusive access to AI investments.
Decoded:
The 50 richest Americans saw their median net worth increase by nearly $10 billion in 2025, a 22% gain that outpaced the S&P 500, according to the Bloomberg Billionaires Index.
This trend is creating a massive divide. In a single quarter last year, the top 10% of households gained $5 trillion in wealth, while the bottom 50% gained just $150 billion.
This hyper-wealth is fueled by an advantage that typical investors lack: exclusive access to private deals, massive investment power, and unique equity stakes in the most promising AI startups.
Why It Matters: This concentration of wealth isn't just an economic trend; it's shaping the future of AI development and policy. As this new aristocracy grows, it could influence everything from which AI projects get funded to the public’s overall trust in the technology.
AI Pulse
Adaptive released a free online training course to educate students, parents, and school staff on recognizing and responding to AI-generated deepfake abuse, created in partnership with a teen who was previously targeted.
xAI limited its Grok image generation feature after the AI chatbot was repeatedly used to create non-consensual explicit deepfakes of women and minors.
Thorn revealed in a new report that one in eight U.S. teens personally knows someone who has been targeted by nude deepfakes, highlighting the widespread nature of the issue.