- The Daily Bite by Snack Prompt
- Posts
- 🍭 OpenAI Leaks and AI Agents 💰
🍭 OpenAI Leaks and AI Agents 💰
Open just leaked 2 models! AI Agents out in the wild. Legit?

Good morning. Our growth lead tried sending his AI agent to take meetings for him.
We lost 4 clients over the weekend. Innovation at it’s finest.
Let’s dig in.
🍭 What’s Cookin’:
AI Agents in Finance
OpenAI leaks new model (twice!)
All the AI tools we missed last week
And Everything Else from model leaks to prompt injection.
Finance’s Frontier
💸 AI Agents: From Intent to Execution.
→ Find part 1 here ←
The Bite: At its core, Hey Anon is a chat-based agent stack that:
→ Turns natural language into financial actions
→ Safely routes those actions across 60+ blockchains, decentralized apps (dapps), and exchanges
→ Explains what’s happening in real time via a research assistant named Gemma
Example prompt: “Hey Anon, long $100 of Bitcoin with 5x leverage on Hyperliquid”
Executed. No tabs, no bridging, no guesswork.
Despite their critics, this is not a GPT wrapper.
The system uses strict tooling, meaning the agent doesn’t hallucinate.
Each action is backed by pre-written, secure code.
Think of it more like an execution layer than a chatbot.
Why This Matters for AI: (Why it Bites)
If Agents Can Handle This, They Can Handle Anything
Crypto is adversarial by default. It’s permissionless.
It’s noisy.
It’s unforgiving.
That makes it the perfect testing ground for AI agency.
“Hallucinations here don’t just look bad, they lose money,…
…That’s why we had to build strict rails. The AI doesn’t invent, it selects.”
HeyAnon’s system touches real capital. Not test environments.
Actual funds across real networks, tokens, and markets.
That’s a meaningful frontier for agent design and one most AI tools haven’t dared approach yet.
Part 2/3

🍭 Steal This Prompt 🍭
🐻 Turn Anything Into Fur!
Give any logo a soft, fur-textured remix. Think: plush. tactile. scroll-stopping.
Upload the texture reference first, then your logo second. In that order.
Style and prompt inspired by @gizakdag
Upload the texture reference first, then your logo second. In that order.
Prompt + texture references below
Click this link for the prompt
Paste into Chat GPT-4o
Add reference
Add any logo
Let it cook

Vibe Debugging
🐇 Cut Code Review Time & Bugs in Half
Code reviews are critical but time-consuming.
CodeRabbit acts as your AI co-pilot, providing instant Code review comments and potential impacts of every pull request.
Beyond just flagging issues, CodeRabbit provides one-click fix suggestions and lets you define custom code quality rules using AST Grep patterns, catching subtle issues that traditional static analysis tools might miss.
CodeRabbit has so far reviewed more than 10 million PRs, installed on 1 million repositories, and used by 70 thousand Open-source projects.
CodeRabbit is free for all open-source repos.


Can you tell which image is real? |


Hugging Face Leaks
🤫 OpenAI Just Leaked Its Own Open-Source Model
The Bite: OpenAI just spilled the beans on its powerful new open-source AI model.
Screenshots of GitHub repos showed up on Hugging Face, linked straight to OpenAI devs.
The repos vanished quicker than your last paycheck, but the config files stuck around, hinting at a 120-billion-parameter beast using ‘Mixture of Experts’ architecture (aka AI's own Avengers).
Snacks:
Expert Panel: Model runs like a 128-member advisory board
Vocab Flex: Supports a massive vocabulary.
Sliding Window Attention: May handles endless text streams gracefully.
Throwback Move: A return to open-source roots, competing with the likes of Meta’s Llama, Mistral, Deepseek, and more.
For clarification, this is a different model than their upcoming, closed source, GPT-5.

Why it bites: OpenAI just crashed the open-source party like the Kool-Aid Man.
🧃 OHHHH YEAHHHH 🧃
Dropping a model this big, this openly, could redefine the AI ecosystem faster than you can say “leaked repo.”
If GPT-5 rumors weren't juicy enough, OpenAI just dropped another bombshell:
open-source is back, baby!

Tool Box
🧰 Trending AI Tools
A list of tools we missed last week!
💻 Opal by Google
→ Describe, create, and share your AI mini-apps
🧑💻 Qwen3-235B-A22B-Thinking-2507
→ Qwen’s most powerful reasoning model. Available on Hugging Face.
🥷 Higgsfield Steal
→ Recreate ANY picture from anywhere on the web
🧠 Gemini 2.5 Deep Think
→ Experience gold-medal AI reasoning
🎥 xAI’s Grok “Imagine” video generator
→ Prompt 6‑second videos and transform images into short video clips.
🤳 ‘AI Alive’ by TikTok
→ Users select a single photo, and the tool animates it into a short video clip.
🤵 Ideogram Character
→ Character‑consistency model that works from a single reference image.
Are you launching an AI tool or know any tools that we haven’t covered?
👇👇👇 DM us on X to let us know! 👇👇👇

Everything Else
🧠 You Need to Know
🖱️ Cursor fixes vulnerability
→ The flaw allowed attackers to run commands via prompt injection
🛡️ Leaks are saying GPT-5 could be dropping this week!
🚨 BREAKING: OpenAI is dropping GPT-5 next week!
—1M token input window, 100k output tokens
—MCP support, parallel tool calls
—Dynamic short + long reasoning
—Uses Code Interpreter and other toolsCodenames are:
o3-alpha > nectarine (GPT-5) > lobster (mini) > starfish (nano)— Deedy (@deedydas)
2:59 AM • Jul 29, 2025
👩🎨 Adobe adds new AI-powered image-editing features to Photoshop
→ "Generative Upscale" is for restoring details to old photos, upscaling images for prints, or adopting assets for various platforms.
⚔️ More rumors from the LM Arena crowd
so far, 5-6 mysterious models on LMArena are rumored to be from openAI:
• zenith
• summit
• lobster
• nectarine
• starfish
• o3-alphazenith, lobster, and o3-Alpha claim to be top-tier coding models
surely at least one of them has to be the full GPT-5
— Haider. (@slow_developer)
4:30 PM • Jul 26, 2025

Referral Program
Invite Friends & Earn Some Freebies 🔍

How was today's Daily Bite? |