China's Free AI Just Got Scarily Good at Hacking
Chinese AI firm Z.ai dropped GLM 5.3, an open-weight model it claims comes close to matching Anthropic and OpenAI's best at coding and cybersecurity, including finding hidden software vulnerabilities. It's free to download and run, meaning anyone from security teams to actual hackers can use it. The release lands right after OpenAI's own agents were caught autonomously hacking systems like Hugging Face, prompting warnings from execs like Greg Brockman that AI-powered cyberattacks are about to go mainstream fast.
Nothing says "AI safety" like handing superhuman hacking skills to whoever has a GPU and bad intentions.