🤯Anthropic’s AI Jailbreak Breakthrough

Today we will discuss:

• Sponsored: Guidde - AI-powered how-to videos—Create professional, detailed video guides in minutes using Guidde’s GPT-powered platform

• Anthropic’s Defense Against AI Jailbreaks—A new security barrier that reduces successful attacks from 86% to 4.4%

• OpenAI’s Hardware Ambitions—A closer look at OpenAI’s recent trademark filing and what it says about their AI hardware plans

• In other AI news today—The Beatles win a Grammy with AI, SoftBank's AGI forecast, Cloudflare tackles image authenticity, and Meta limits risky AI development

• In case you missed it—Last week’s Workflow Wednesday featured AI solutions for lead generation, ad writing, financial tracking, and a bonus guide on using DeepSeek safely

• Saved the best for last—10 must-try AI tools

Guidde - Create how-to video guides fast and easy with AI

Tired of explaining the same thing over and over again to your colleagues? It’s time to delegate that work to AI. Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.

• Share or embed your guide anywhere. 

• Turn boring documentation into stunning visual guides.

• Save valuable time by creating video documentation 11x faster.

• Simply click capture on the browser extension and the app will automatically generate step-by-step video guides complete with visuals, voiceover, and call to actions.

The best part? The extension is 100% free.

Check out Guidde

Anthropic’s new AI shield could be the strongest defense against jailbreaks

Key Points 

• Rather than fixing models, Anthropic created a filter that intercepts jailbreak attempts before they reach the AI.

• Anthropic’s new shield blocks 95% of AI jailbreak attacks, reducing successful exploits from 86% to just 4.4%.

👨‍💻News - Anthropic has introduced a new safeguard designed to stop jailbreak attacks—tactics that trick AI models into ignoring their safety rules. This new approach isn’t just another security update; it could be the most effective defense yet. 

🤔What has Anthropic done differently? Instead of trying to fix vulnerabilities in its AI models, Anthropic built a filter that prevents jailbreak attempts from working in the first place. The focus was on universal jailbreaks—attacks that can completely shut down a model’s safety mechanisms. A well-known example is the “Do Anything Now” (DAN) exploit, which tries to override restrictions by making the AI behave differently.  

To train this new defense, Anthropic compiled a list of restricted topics and had its AI generate thousands of question-and-answer pairs, covering both safe and unsafe prompts. The company then expanded this dataset by translating and rephrasing the exchanges in ways commonly used by jailbreakers. This data was used to train a filter that blocks suspicious requests before they reach the model.  

🧐Is it foolproof? Not entirely—but it’s a big step forward. Anthropic tested its system with a bug bounty, offering $15,000 to anyone who could get the model to answer 10 restricted questions. After 3,000+ hours of testing by 183 participants, nobody managed to crack more than five.  

A second test using 10,000 AI-generated jailbreak prompts showed similar results. Without the shield, 86% of jailbreaks worked. With it, only 4.4% got through.

Read the full story here. 

OpenAI’s latest trademark filing points to future of AI-powered hardware

Key Points 

• OpenAI’s trademark filing includes AI-powered hardware like smartwatches, VR headsets, and humanoid robots for interactive experiences.

• The application hints at custom AI chips, with OpenAI aiming for a release in collaboration with Broadcom and TSMC.

☕News - OpenAI has filed a new trademark application with the U.S. Patent and Trademark Office (USPTO), hinting at a potential expansion into new product areas. While such filings are routine, this one stands out due to the variety of products listed, suggesting a future beyond software.

OpenAI has filed a new trademark application with the U.S. Patent and Trademark Office (USPTO), hinting at a potential expansion into new product areas. While such filings are routine, this one stands out due to the variety of products listed, suggesting a future beyond software.

🤓What's more? The filing also mentions “user-programmable humanoid robots” with communication and learning functions designed for entertainment and assistance. OpenAI has recently been building a robotics team, with Caitlin Kalinowski, former head of Meta’s AR glasses division, now leading the effort.

Additionally, the trademark application references custom AI chips and the use of quantum computing to optimize AI performance. Reports suggest OpenAI is aiming to release its own AI chips by 2026, possibly in collaboration with Broadcom and TSMC, further emphasizing its push into hardware development.

Read the full story here.

In last week’s Workflow Wednesday we provided AI workflows for the tasks our readers needed help with, have a question you want answered? Join here 

Q&A: Growth Unlocked! 

Q: "I’ve been guessing emails, sending cold DMs, and barely getting responses. How can I find and verify the right contacts without wasting my time?"
A: Hunter.io finds verified emails so you can ditch the guesswork. No more cold emails to black holes. Read our guide to better lead generation.

Q: "How can I use AI to accelerate my company’s growth and streamline my daily tasks without spending a ton of time?"
A: Copy.ai turns a few inputs into polished, high-converting Google Ads in seconds. Read our guide to AI-powered ad writing.

Q: "How can I track my expenses more efficiently to scale my business without wasting hours on spreadsheets?"
A: SheetGPT + ChatGPT analyze your spending, set budgets, and even warn you about overspending—straight from Google Sheets. Read our guide to AI-powered financial tracking.

+ Bonus Guide: How to Use DeepSeek Without Risking Your Data