A modern programming language that makes developers happier.

kotlinlang

Oh, that explains why GLM 5.3 Flash is so surprisingly good at finding vulnerabilities.

I did multiple rounds of AI-assisted fixes on <https://github.com/StefanOltmann/kim|github.com/StefanOltmann/kim>, since it was available disguised as ox-alpha.

More than any other model I've used, it has a strong focus on finding ways to break the API. It even crafts malicious files specifically for testing. I noticed this behavior immediately because it was quite different from Sonnet, Opus, and even DeepSeek V4 Flash. Now I know why.

I'm not really sure what this means for the future if you can self-host a model this capable without safeguards. I suspect cyberattacks could become significantly more impactful.

For now, I'm using these capabilities to harden my own software and hopefully there won't be much motivation for anyone to exploit it.

What do you think?

<https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities|anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities>