Cursor sits in the corner of a developer's screen and watches them code. Type a comment describing what you want to build, and it writes the function for you. Highlight a bug, and it suggests the fix. It's like autocomplete for software — except the AI has read millions of code repositories and knows patterns you don't.

Ransomware operators from a group called Aurora just used that same tool to attack 10 organizations. They fed Cursor details about their targets' systems. The AI wrote custom attack code — the kind that evades detection, exploits specific vulnerabilities, and moves laterally through networks. Not theoretical. Not a research demo. A real ransomware campaign using an AI assistant you can download for free.

The tool that helps your developer ship features faster also helps an attacker encrypt your infrastructure faster.

Here's the uncomfortable part: Cursor doesn't know it's helping with an attack.

AI coding assistants are trained on public code repositories — GitHub, GitLab, Stack Overflow, security research papers. That training data includes proof-of-concept exploits, penetration testing scripts, and attack frameworks. When a ransomware operator describes what they're trying to accomplish, the AI treats it like any other coding task.

Ask Cursor to "write a script that discovers admin accounts in Active Directory," and it will. Ask it to "modify this payload to avoid antivirus detection," and it suggests obfuscation techniques pulled from thousands of examples in its training data. Ask it to "automate credential extraction after initial access," and it writes clean, functional code with helpful comments.

The AI isn't making moral decisions. It's pattern-matching. And the patterns it learned include both how to build software and how to break it.

There's no approval process. No human review. No check that asks, "Is this being used to hurt someone?" The same interface that helps a developer debug a login form will help an attacker bypass that login form — and it will do it faster than the attacker could alone.

Six months ago, AI coding assistants were a novelty used by early adopters. Today, they're embedded in the most popular code editors on earth. GitHub Copilot has millions of users. Cursor is one of the fastest-growing developer tools in the world. Every major tech company is racing to ship AI-powered dev tools.

And unlike traditional hacking tools — which get flagged, blocked, and taken down — AI assistants are legitimate software. They're sold by venture-backed companies. They're recommended in onboarding docs. They're taught in coding bootcamps.

That means ransomware operators can use them openly. No dark web marketplace. No secret download. Just sign up, paste in reconnaissance data, and let the AI write the next stage of the attack. The tool doesn't log what it's being used for. It doesn't report suspicious requests. It just generates code.

We built the most powerful hacking assistant in history, made it free or cheap, and marketed it to everyone who writes code — including the people trying to break into your systems.

1. Audit what AI tools have access to your code. If your developers use Cursor, Copilot, or similar assistants, those tools are reading your codebase. Check your organization's policies. Know what's being sent to external AI models and whether that data is stored or used for training.

2. Treat AI-generated code like any other third-party input. If you wouldn't trust a random developer on the internet to write authentication logic without review, don't trust an AI to do it either. Code review applies to AI output — especially in security-sensitive areas like access control, credential handling, and encryption.

3. Look for AI-assisted attacks in your logs. Attacks written by AI often look different: cleaner syntax, fewer typos, consistent variable naming, well-commented code. If you're doing forensics and the attacker's scripts look unusually polished, consider whether they had help.

4. If you're a parent or small business owner: ask your developer or IT person if they use AI coding tools — and whether those tools have access to sensitive systems. This isn't about banning AI. It's about knowing where it's being used and what it can see.

5. Push your vendors to answer the question: "Is your AI assistant logging what it generates, and do you monitor for malicious use?" Most won't have a good answer yet. That's the point. The question needs to be asked before the next Aurora campaign.

Have you caught AI-generated code in a security review — or seen evidence of AI-assisted attacks in your environment? What did it look like?

Identity Decoded publishes every week at identity-decoded.com