Free AI Model Now Writes Code and Finds Security Flaws

A team called Z.ai just released GLM-5.3, the newest version of their open AI model. Unlike some AI upgrades that need new training behind closed doors, this one uses the same base model as before. The improvements come from post-training — imagine giving an already smart assistant extra coaching sessions until it becomes an expert at specific tasks. The result? A model that's much better at complex coding tasks, like building multi-step software programs without help.

Here's what makes this useful for you: GLM-5.3 is free and open, meaning any developer can download it and build tools on top of it. That could lead to cheaper and faster apps, better home automation, and even junior programmers getting AI help that works like a senior engineer. On internal tests, it was 50% better than the previous version at coding challenges, and it scores higher than other open models on public tests that measure real-world software tasks.

But there's a darker side. The team says they noticed a surprise: the model became very good at cyber tasks like finding security holes and exploiting them. In fact, it more than doubled the previous version's ability to break into systems after discovering a vulnerability. That means the same free tool that helps people write secure software can also help attackers find ways in. This is a classic dual-use problem — a powerful tool that can be used for good or harm, and it's now in everyone's hands.

What should you take away? AI is no longer just a chatbot that writes emails. It's becoming a practical worker that can build software and check for weaknesses. For businesses, this could cut development costs and improve security. For the rest of us, it means the AI playing field is opening up — but so is the potential for more sophisticated cyberattacks. The future of work and security will likely revolve around who uses these tools better.