# Vercel — Security boundaries in agentic architectures

- Company: Vercel (vercel.com)
- Announced: 2026-02-24T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/security-boundaries-in-agentic-architectures
- Record: https://forck.live/items/20744-security-boundaries-in-agentic-architectures
- Subject: v0 / AI SDK / AI Gateway

Vercel published a guide on security boundaries in agentic architectures, recommending that teams separate the security contexts of the agent, its secrets, generated code execution, and the filesystem. The post identifies four actors in an agentic system with different trust levels and argues that running all components in a single context enables prompt injection attacks that can exfiltrate credentials.

## Evidence

Verbatim from https://vercel.com/blog/security-boundaries-in-agentic-architectures:

> While most teams run all of these components in a single security context, because that's how the default tooling works, we recommend thinking about these security boundaries differently.

---

Record: https://forck.live/items/20744-security-boundaries-in-agentic-architectures
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md