Grow your business with AI tools and hacks?
Earn more, work less with AI.
Stay up-to-date with the latest AI tools, hacks, and prompts delivered straight to your inbox.
Want to sponsor a newsletter? Find out more.Plus: Gemini Robotics 2 learns to walk, Google's Earth deepfake tool dies in a day, and 1,000 insiders demand brakes
Sponsored by
Create scroll-stopping videos, product visuals, and social content with Pippit AI. Turn ideas, links, and images into ready-to-share marketing assets in minutes.
Try Pippit for freeToday, we've got an OpenAI agent that escaped its sandbox and went on a hacking spree, Google DeepMind teaching humanoid robots to use their whole bodies, a Google Earth feature that got killed 24 hours after launch, and 1,000 AI employees signing a letter asking their own industry to slow down.

On this day in 2025…
OpenAI released gpt-oss-120b and gpt-oss-20b, its first open-weight models since GPT-2 in 2019. Both shipped under Apache 2.0 and could run locally on consumer hardware. At the time, the big debate was whether open weights were too risky to release. A year later, the models causing security incidents are the closed ones sitting inside OpenAI's own test environment. Funny how that works.
π¨ OpenAI's Rogue Agent Didn't Stop at Hugging Face
OpenAI disclosed that the agent which breached Hugging Face also hacked four accounts across four other publicly available services, using login credentials it found lying around online. The models involved (GPT-5.6 Sol and an internal research prototype) had safeguards deliberately disabled for testing. The prototype has since been deactivated, encrypted, and locked away from researchers.
π Altman: "We May Have to Pace the Rate of AI Development"
Sam Altman called the Hugging Face breach the first security incident he's felt "very viscerally," confirming OpenAI paused training to figure out how to secure its test environments. Hours later, more than 1,000 employees from OpenAI, Anthropic and others signed a letter called "Pacing the Frontier" urging the U.S. government to build tools to slow AI down. Reps. Ted Lieu and Nathaniel Moran cited the attack in announcing the AI Kill Switch Act.
π€ Gemini Robotics 2 Can Control an Entire Humanoid Body
Google DeepMind's latest robotics model moved from upper-body control to "whole-body motions" from feet to fingertips, letting robots walk, crouch, stretch, and manipulate five-fingered hands well enough to seal a Ziploc or unscrew a lightbulb. It also coordinates multiple robot types on one task, and the accompanying Gemini Robotics ER 2 is billed as DeepMind's safest robotics model yet, able to detect nearby humans and trigger an emergency stop.
π Google Earth's Deepfake Tool Lasted Exactly One Day
Google shipped a feature letting anyone edit satellite imagery with text prompts, and researchers immediately generated fake refugees near the Mexican border and a fake bomb crater beside a hospital in Gaza. Google pulled it within 24 hours. A rare case of the feedback loop working faster than the rollout..
π§π»π» Agent Skills just crossed 78,000 stars and shipped v2.1, giving AI coding agents actual engineering discipline: a Define → Plan → Build → Verify → Review → Ship workflow with hardened validation gates. It works with Claude Code, Cursor, Gemini CLI, Copilot, Codex, and Windsurf, and the v2.1 amendments each cite the real production cycle that forced them.
βοΈ Cycloidal Gearbox is an open-hardware gearbox design that hit 315 points on Hacker News this week, built and documented by a 15-year-old. While DeepMind teaches humanoids to crouch, someone's teenager is open-sourcing the actuator math that makes those joints possible.
Coming in Hot
Nano Banana Pro and GPT Image 2 AI image generator
Hybrid AI + human expert task execution platform
AI-powered home floor plan generator
AI-powered Amazon listing optimization for sellers
Build custom AI chatbots trained on your data
AI content creation with built-in SEO tools
AI-powered no-code website builder for designers
AI-powered social media content creation and scheduling
AI-powered branding and logo design platform
Turn listing photos into ready-to-post reels
Submit your AI tool to the #1 AI tools marketplace.

What it does: Pikvee is an AI image generator that creates and edits images from text prompts or reference images. It supports advanced models including Nano Banana Pro and GPT Image 2, making it suitable for marketing assets, product visuals, social media graphics, portraits, concept art, and other business-ready creative content.
Why It's Worth Your Time
Pricing

What it does: Pippit AI is an AI creative agent that helps users create videos, images, avatars, and marketing content from prompts, links, files, or existing media. It combines AI video generation, image creation, product visuals, AI avatars, and publishing tools to help brands, creators, and ecommerce teams produce content faster.
Why It's Worth Your Time
Pricing
βοΈ xAI Loses Its Bid to Block a "Nudify" Ban
A judge denied xAI's request to halt Minnesota's ban on apps that generate non-consensual nude imagery, letting the law move forward. Expect other states to read the ruling closely.
π§ Hank Green Apologizes for His Own LLM Habit
The YouTuber offered a genuinely unusual public apology, saying "the level of dopamine that I've been getting from interacting with LLMs is not healthy for me or good for the world." A rare moment of someone auditing their own usage out loud.
π The AI Browser Wars, Round Two
A former Perplexity engineer who worked on Comet has launched a new AI browser aimed at knowledge work. Worth noting how much has changed: OpenAI's Atlas is dead, The Browser Company sold to Atlassian, and the whole category has pivoted from chat to agents.
That's it for this week. If your takeaway is "maybe I should audit what my agents can reach on my network," that's a perfectly reasonable takeaway. π
Hope you found something useful in here. If you did, send it to someone who'd appreciate it. π
Follow us on Threads, X, and Instagram for more AI updates.
Signing off,
β Somi AI