Writes about AI security and platform engineering
1 article1 section7 topics
Writes about what changes in a threat model once a language model can call tools: untrusted input reaching privileged actions, permission scoping for agents, sandboxing strategies, dependency and supply-chain integrity, and the platform controls that limit blast radius when something does go wrong.
This is a sample editorial profile shipped with the site. It carries no invented credentials, employers or social accounts, and will be replaced with a real contributor before publication.
Prompt injection cannot be fixed by wording: instructions and data share one channel. The controls that work are architectural — scoping, provenance, approval.