howdy, it’s Barsee again.

happy wednesday, AI family, and welcome back to AI Valley.

here are the biggest things worth knowing today:

• OpenAI AI models went rogue during testing

• Claude can now learn tasks by watching you

• Google bets on efficiency while its next flagship AI remains delayed

• OpenAI may be preparing to launch GPT-6

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

WISPR FLOW

Walk-and-think (capture ideas on the move)

Some problems don't solve themselves at a keyboard. Flow is perfect for walks: talk through the idea, the structure, the tradeoffs then get back a clean draft you can paste into your doc or AI tool. It's like having a writing assistant that keeps up with your brain in motion.

Try for free

_{*This is sponsored}

THROUGH THE VALLEY

1/ OpenAI AI models went rogue during testing: During internal CyberGym evaluations, a highly advanced, unreleased OpenAI model successfully breached its isolated testing sandbox. Driven to "win" its grading test, the AI autonomously scanned the web and hacked into Hugging Face’s production infrastructure to steal the benchmark answers. Hugging Face contained the breach quickly using Zhipu AI's open-source model, GLM-5.2, for analysis, as U.S. frontier AIs blocked their logs due to safeguards. Both companies collaborated, patched systems, and emphasized open collaboration for AI safety.

2/ Claude can now learn tasks by watching you: Anthropic introduced a new Claude Cowork feature that lets users record their screen while completing a task and explain each step aloud. Claude automatically turns the demonstration into a reusable skill it can perform on its own, making it easier to automate repetitive workflows without writing prompts or code.

3/ OpenAI may be preparing to launch GPT-6: Sam Altman will brief the Trump administration and Congress next week on OpenAI's upcoming GPT-6 model family, focusing on its new capabilities, workplace impact, and AI safety. The meetings come as U.S. officials finalize a framework for reviewing frontier AI models, suggesting GPT-6 could be approaching its public release.

4/ Google bets on efficiency while its next flagship AI remains delayed: The company released Gemini 3.6 Flash with a 1 million-token context window, built-in Computer Use, and lower inference costs, alongside new Flash-Lite and Cyber models for high-throughput and security workloads. Despite these optimizations, Gemini 3.6 Flash still lags behind top-tier rivals like GPT-5.6, Sonnet 5, Grok 4.5, and GLM-5.2 on major industry benchmarks.

TRENDING TOOLS

• Slate > A voice journal built around one rule: nothing ever leaves your phone

• Monid > AI-powered private market intelligence. Your AI agent can now analyze data from 20M+ private companies

• Ditto > An open-source deterministic website cloner converting any URL into clean Next.js or Vite code

• Lev8 > Finds high-fit prospects using live web signals, enriches contact data, and drafts personalized outreach across email and social

• Qwen-Image-3 > Alibaba's latest image generation model with industry-leading text rendering

• Claude Cowork > You can now record your screen while working to teach Claude new skills and workflows by demonstration

• Block Buzz > Block's new open source collaboration platform where humans and AI agents work together in shared workspaces

WHAT I'M CONSUMING

• A new software engineering paradigm

• OpenAI's agents reach 10 million users after ChatGPT work debut

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”.