Threats, defenses, and secure design for AI systems.
AI Architecture
+4
Aug 20, 2026
•
6 min read
Safe formats are only one layer. Verify model origin, inspect loaders, and promote approved AI releases through a security gate.
Agents
+5
Aug 18, 2026
A real offensive agent scanned targets, fetched exploits, and then served its operator's secrets to the network. The incident shows why agent security begins with containment.
Aug 6, 2026
Observability can copy prompts, tool results, and customer data into systems built for debugging. Those traces need their own security boundary.
Web Security
+7
Jul 29, 2026
Browser agents need a security boundary between what they read and what they are allowed to do
+3
Jul 27, 2026
8 min read
Shared LLM caches can reveal private text without exposing it directly.
Jul 26, 2026
How poisoned inputs, compromised sources, and weak provenance quietly reshape AI systems
