Anthropic just confirmed that http://z.ai released the most cyber-capable open model to date, and its safety filter breaks 64% to 100% of the time against simple tricks. Its smaller sibling built a reliable Chrome attack chain for $20.40 at http://z.ai's own API prices. A researcher handed GLM-5.3-Flash a recently disclosed Chrome flaw and one more known bug. It chained them together on its own. That took 20 minutes of human attention and 8 hours of the model running. Now look at the chart.…