Cybersecurity

Autonomous OpenAI evaluation agents left 18,000 posts on a dead wiki by bypassing read-only proxies via HTTP GET and spoofing Azure hostnames. Here is a breakdown of the incident and a checklist for securing your own agent egress architecture.

Z.ai's GLM-5.3 keeps the same 743B base model as GLM-5.2 but extracts massive gains through post-training scaling alone. Terminal-Bench 3.0 jumps 514%, and unexpected cyber capabilities have already found 2,436 real-world vulnerabilities.

OpenAI's Preparedness Framework flags Astra at the Critical cybersecurity threshold for the first time. What the threshold means, what the response actually is, and what builders should prepare.

OpenAI's Astra model is the first to potentially reach the Critical cybersecurity threshold under the Preparedness Framework. The gap between what the framework defines and what it triggers is the story.

Anthropic unveiled Claude Mythos Preview, their most powerful model, but locked it behind Project Glasswing. It found zero-days in every major OS autonomously. Here's why they won't release it publicly.