Anthropic Builds a Hacker Hunter, a Video Model Nails #1, and Your Screen Gets a Tutor

 🔥🔥🔥 Three Highly Curated AI Stories

1. Anthropic launches Claude Mythos Preview and Project Glasswing

Anthropic's newest frontier model is here, and it comes with a mission. Claude Mythos Preview is the engine behind Project Glasswing, an urgent initiative focused on securing the world's most critical software.

What makes Mythos different is where it's pointed. The model is specifically built to identify software vulnerabilities, and Anthropic says it performs better than all but the most skilled human security researchers. That's not a casual benchmark — most enterprise security teams would pay serious money for that capability.

Project Glasswing is the broader wrapper around this: using AI to get ahead of attackers rather than just reacting to them. It's Anthropic planting a flag in one of the highest-stakes corners of the AI space.

2. Claude Code's /autofix-pr now runs straight from the command line

Claude Code users, this one's for you. The /autofix-pr command can now be triggered directly from your terminal, skipping the context-switch back to the UI.

Finish a PR, run /autofix-pr, and it sends your full session to the cloud. The autofixer gets complete context to address CI failures and reviewer comments without you having to explain anything.

It's a small surface change with a real productivity payoff — exactly the kind of thing you don't notice until it's gone.

3. HappyHorse-1.0 drops at number one on the video leaderboard

A new video model called HappyHorse-1.0 entered the scene and immediately claimed the top spot, leading in both text-to-video and image-to-video. The name is a lot. The performance apparently backs it up.

Early testing shows it handles multi-shot videos and detailed prompt instructions better than most models in the space — both of which have historically been weak spots for AI video generation.

No public release timeline yet, but it's already one to watch.

🔥🔥 Two AI Tools To Try Today

1. 📡 Exa Monitors

Exa is a search engine built from scratch specifically for AI agents — not the kind designed for humans typing into a search bar. Today they launched Monitors, which gives your agent a live radar for the web. You define what to find and how often, and Monitors surfaces new matching results automatically via webhook.

If you're building agents that need to track competitor moves, job postings, research updates, or any kind of real-time signal, this closes a gap that most search APIs still leave wide open.

P.S. We wrote a full guide on using Claude and Exa together for LinkedIn lead generation. If you want to see what this combo can actually do in practice, worth bookmarking

2. Clicky

Clicky is an AI teacher that lives next to your cursor. It can see your screen, speak to you in real time, and literally point at elements on the page — like a real tutor sitting beside you while you work.

The creator built it while learning Davinci Resolve and gave it a 10/10 rating after a few days of use. If you're picking up any new tool or software, this is a genuinely different approach to learning than watching a YouTube tutorial on mute.

🔥 Things You Should Know About AI

Turn Claude and Exa into a LinkedIn lead generation machine

Most people use Claude to write things. Fewer people know you can wire it to a web search engine built for agents and automate a big chunk of your lead gen workflow. Here's how:

1. Sign up for Exa and grab your API key from the dashboard.

2. Write a clear lead profile in Claude: target role, industry, company size, and any keywords that define your ideal contact.

3. Feed those criteria into Exa's search to pull a list of matching LinkedIn profiles.

4. Ask Claude to draft a personalized outreach message for each lead using the profile details Exa returns.

5. Review the messages, make small tweaks, and send.

Full step-by-step walkthrough here.