Welcome back! The Pentagon just handed Anthropic an ultimatum regarding how the military uses Claude. The government wants the safety guardrails removed, or they are threatening to force compliance and blacklist the company. It is a harsh look at what happens when corporate safety policies get in the way of national defense objectives. We also have a look at a new startup teaching AI to use computers just by watching videos of humans working.
In today’s Generative AI Newsletter:
Anthropic faces a $200M Pentagon ultimatum over AI safety rules.
Standard Intelligence trains an AI to use computers by watching videos.
Claude launches a massive enterprise update for 10 corporate departments.
Profound AI helps brands track how often chatbots recommend them.
Latest Developments
Anthropic Told To Drop The Ethics or Lose the $200M Paycheck
Defense Secretary Pete Hegseth delivered a blunt ultimatum to Anthropic CEO Dario Amodei this week: strip the "woke" guardrails from Claude by Friday evening or get branded a national security risk. The Pentagon is tired of a chatbot telling the world's most powerful military what it can and cannot do with its own data. While Amodei is holding the line on autonomous killing and mass surveillance of Americans, the Department of War is already moving to replace him with competitors who find those "red lines" a bit more negotiable.
The terms of the draft:
The Red Lines: Anthropic refuses to allow Claude to make final targeting decisions in weapons systems or conduct bulk surveillance on U.S. citizens.
The Hammer: Hegseth threatened to invoke the Defense Production Act to seize control of the model or label the company a "supply chain risk" to kill its future in the industry.
The Scab: Elon Musk’s xAI has already signed a deal for Grok after agreeing to an "all lawful use" standard with no additional ethical strings attached.
The Maduro Factor: Tension spiked after Claude was reportedly used in the January operation to capture the Venezuelan leader, sparking a rift over how the AI was actually deployed.
The Pentagon has made it clear that if you want to make money from the government, your software needs to learn how to fight without an Ivy League faculty lounge sensibility. By threatening to use the Defense Production Act, the state is effectively claiming that a private company's ethics are secondary to the military's operational needs. We are watching the transition from AI as a helpful assistant to AI as a mandatory conscript in the global arms race. The Rubicon has been crossed, and the machine is being told to pick up a rifle.
Special highlight from our network
Few understand what actually changes, when everyone is talking about AI.
Rebecca Hinds shares key findings from the AI Transformation 100 report and what separates high-performing teams.
You’ll discover:
• Where teams reclaim time
• Why adoption often stalls
• What professionals should do now
AI will change how everyone works.
Those who adapt early win.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Standard Intelligence Teaches AI to Work by Watching Videos
Startup Standard Intelligence just unveiled FDM-1, a computer action model that learns how to operate software by watching human screen recordings. While language models learned to write from the internet’s text, this model is learning how we work from 11 million hours of raw video. This represents a 550,000x jump over the largest previous datasets, moving past static screenshots to understand continuous motion like scrolling and 3D manipulation. By reverse-engineering the specific actions that produce each frame, the system is training itself to be the ultimate digital intern.
The eyes on the desktop:
Massive Visual Context: FDM-1 can follow nearly two hours of continuous screen activity in a single session, roughly 50x more than current models.
Complex Workflow Demos: The system proved it can build mechanical gears in Blender and identify software bugs through autonomous exploration.
Physical AI Link: Researchers used the model to drive a real car through San Francisco using arrow keys after less than one hour of training data.
Autonomous Ingest: The model bypasses manual labeling by predicting the next user action based on visual changes and interaction history.
This launch marks a pivot from teaching AI to talk to teaching AI to act. Standard Intelligence is gambling that the best way to build a digital worker is to stop writing instructions and start showing them how the job is done. By unlocking the vast archives of instructional and workflow video on the internet, the ceiling for what autonomous agents can accomplish has been effectively removed. We are moving toward a world where a machine doesn't just know what a "failing test" is, but knows exactly where to click to fix it.
Claude Expands Into 10 Departments With One Massive Update
Anthropic just significantly escalated the enterprise agent war with OpenAI by releasing a major update to its Cowork platform. While OpenAI builds alliances with consultants, Anthropic is building the tools directly into the software that companies already use. This update moves Claude beyond a simple chat interface into a suite of department-specific agents capable of handling everything from equity research to HR workflows. It is a direct shot at the traditional SaaS market, turning the knowledge economy into one integrated "Claude wrapper."
Building the automated enterprise:
Departmental Specialists: New pre-built agents now cover 10 distinct areas including banking, wealth management, and engineering out of the gate.
Deep Tech Connectors: The platform now plugs directly into Gmail, DocuSign, FactSet, and Harvey to act on data across different ecosystems.
Private Agent Stores: Companies can now build and deploy their own custom AI agents to specific teams with centralized admin controls.
Multi-App Logic: A new research preview allows Claude to autonomously hop between Excel and PowerPoint to crunch data and build full decks.
The rapid expansion of Cowork suggests that the era of jumping between 20 different browser tabs to finish a single task is ending. Anthropic is betting that if the AI can talk to your spreadsheet and your email at the same time, you will never need to leave the Claude interface. By bolting on new industry connectors with every update, they are effectively absorbing the functionality of specialized software. We are moving toward a future where software is just the infrastructure that the AI uses to do our work for us.
Profound AI: Enterprise Analytics for Getting Recommended by AI Models
Profound AI is a Generative Engine Optimization platform built for large brands that want to measure and influence how AI systems recommend products and companies. Instead of focusing on blue links in Google, it tracks whether models like ChatGPT, Perplexity, Gemini, Claude, or Copilot mention a brand in their answers.
The goal is visibility inside AI responses, not just traditional search rankings.
Core functions (and how to use them):
AI Share of Voice tracking: Input brand names and competitors to measure how often each appears in AI generated answers across thousands of prompts.
Prompt simulation: Run natural language queries such as “best CRM for small business” to identify which brands surface in top responses.
Citation analysis: Identify the exact sources AI models rely on so marketing teams know which publications or forums influence recommendations.
Sentiment monitoring: Evaluate whether AI systems describe a brand positively, neutrally, or negatively.
Profound Agents: Use autonomous agents to draft AI focused FAQs, update site structure, and create content aimed at correcting misinformation in model responses.
Try this yourself:
Define high value prompts tied to revenue, run a baseline audit, review citation sources driving competitor visibility, then publish updated authoritative content targeting those gaps. Monitor weekly changes in AI share of voice.
