We have a year to fix security everywhere

GLM 5.3-flash, an open-weight LLM from Z.ai, runs locally on consumer hardware costing $5–15 k and processes 20–45 tokens per second on a $9.5 k Apple M5 Mac Studio. Ablated variants score 0% on Harmbench-320 and ~84.5% on CyberGym, showing capability to generate malicious code and exploit real-world vulnerabilities. Project Glasswing and Daybreak aim to use such models to locate and remediate security flaws industry-wide within a year.