# Vercel — Building secure AI agents

- Company: Vercel (vercel.com)
- Announced: 2025-06-09T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/building-secure-ai-agents
- Record: https://forck.live/items/20866-building-secure-ai-agents
- Subject: v0 / AI SDK / AI Gateway

Vercel published a guide on building secure AI agents, focusing on prompt injection risks. It recommends assuming total compromise of the prompt, scoping tools to the caller's authority, and treating model output as untrusted. The post also introduces two open-source packages: harden-react-markdown and markdown-to-markdown-sanitizer.

## Evidence

Verbatim from https://vercel.com/blog/building-secure-ai-agents:

> When designing secure AI agents, assume the attacker controls the entire prompt. That includes the original query, any user input, any data retrieved from tools, and any intermediate content passed to the model.

---

Record: https://forck.live/items/20866-building-secure-ai-agents
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md