Howdy, it’s Barsee again.

Happy Wednesday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

• Anthropic accuses 3 Chinese companies of stealing

• Anthropic faces Pentagon pressure over AI safety limits

• Researchers show how easy it is to feed AI false info

• AI learns how to use software by watching screens

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

INNOVATING WITH AI

Case Study: Dan’s 1-Year Transformation To 6-Figure AI Consultant

Dan had no tech background and no business experience – just a love for AI and a hunch it could become something more. Through The AI Consultancy Project, he landed his first clients, found a niche, and built a real business. This case study breaks down his journey – the early stumbles, the system that worked, and how he made the leap.

Read the Full Case Study — Free →

_{*This is sponsored}

THROUGH THE VALLEY

1/ Anthropic accuses 3 Chinese companies of stealing

Anthropic says several Chinese AI labs, including DeepSeek, Moonshot, and MiniMax, used large-scale model distillation to copy Claude’s capabilities without permission. The company claims it uncovered around 24,000 fake accounts that produced more than 16 million interactions to extract Claude’s behavior and train rival models. While distillation is a common research method, Anthropic argues this was done in secret and bypassed safety controls meant to limit misuse, such as cyber or bio risks. OpenAI and Google have raised similar concerns, as open source Chinese models gain traction and powerful capabilities spread without clear safeguards.

2/ Anthropic faces Pentagon pressure over AI safety limits

The Pentagon has reportedly told Anthropic CEO Dario Amodei to loosen Claude’s military safety limits or risk losing U.S. defense contracts. Defense Secretary Pete Hegseth warned that Anthropic could lose its $200 million deal and face pressure under the Defense Production Act if it blocks certain military uses. Anthropic has said it will not support fully autonomous weapons or large-scale surveillance of Americans. Claude was the first AI used inside classified Pentagon systems, but rivals like OpenAI, Google, and xAI are now competing for the same contracts. The standoff highlights growing tension between AI safety policies and national security demands.

3/ Researchers show how easy it is to feed AI false info

A journalist showed how easily AI chatbots and AI search tools can be fooled into repeating false claims planted on a single webpage. He published a fake article about “tech journalists who eat the most hot dogs,” and within a day, major AI systems repeated it as if it were true. This works best in so-called data voids, where few reliable sources exist. The same tactic is now used by marketers and scammers to push biased product lists and misleading advice, including in health and finance. Because AI answers sound confident, people tend to trust them, which makes this kind of manipulation especially dangerous.

4/ AI learns how to use software by watching screens

Standard Intelligence has introduced FDM-1, an AI model trained to use computers by watching screen recordings. The system learned from 11 million hours of video, far more than any open dataset so far, and figures out which actions caused each change on screen. FDM-1 can follow long sessions of real work and has already handled tasks like CAD design, finding software bugs, and even controlling a real car using live inputs through a keyboard. The goal is to teach AI how people actually use software in the real world, instead of relying only on text data.

TRENDING TOOLS

• Rork Max > One-shot almost any app for iPhone and Apple devices, including Watch, TV, and Vision Pro

• Inception Labs Mercury 2 > A diffusion-based LLM built for stronger reasoning

• Interpreter > A desktop AI agent that fills PDFs, edits Excel and Word files, and learns new skills. Runs locally and works with any model

• Google Pomelli > Turn simple product photos into clean, studio-quality images instantly

• Qwen 3.5 > A powerful open Chinese model family that runs frontier-class models fully on your own Mac

• git-lrc > Free, unlimited AI code reviews that run automatically on every commit

• Profound > See exactly what ChatGPT, Perplexity, and Gemini say about your brand

• Devin 2.2 > Writes code, opens a virtual desktop to test it, catches its own bugs, and fixes them before you review

• SkillForge > Turn screen recordings into reusable, agent-ready skills

• Guideless > Create AI-narrated software walkthrough videos in minutes

• Fern > An AI meeting copilot that actively participates and coaches you in real time

• Origami.chat > Find high-quality leads with a single prompt.

WHAT I'M CONSUMING

• THE 2028 global intelligence crisis

• The software industrial revolution

• I let AI agents train their own models. Here's what actually happened

• AI #156 Part 1: They do mean the effect on jobs

• DeepSeek is about to drop a new model, and the Nasdaq is bracing for impact

• You can buy a humanoid robot on Amazon now

• X is testing a new toggle to label your AI-generated slop

• If AI makes human labor obsolete, who decides who gets to eat?

INDUSTRY MOVES

Stripe said to eye potential PayPal acquisition

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.