Analysis
Sharing secrets with LLM agents
Publication date unknown · Discovered vepsala.dev
Page published
Publication date unknown · First observed: 18 Nov 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Coding agents can expose reachable credentials even when secrets never enter the conversation, so keeping keys out of prompts alone does not prevent theft.
“Sharing secrets with LLM agents” examines how coding agents can expose credentials through both their conversation context and their execution environment. The article discusses prompt injection, malicious dependencies, compromised tools and model infrastructure, arguing against running agents directly on a credential-rich main operating system.