#agent-security

2 articles

zenpresa September 12, 2026

Agent Tool Safety: Prompt Injection, Out-of-Band Confirmation Codes, and Idempotency Keys for Money-Moving Tools

How Zenpresa stops its AI assistant being tricked into refunds: a threat model for untrusted text, out-of-band confirmation codes, and idempotent tools.

openai-agents-sdk April 22, 2026

OpenAI Agents SDK in 2026: Sandboxes, MCP, and When to Use It vs the Responses API

When to use the OpenAI Agents SDK vs the Responses API, where MCP fits, when Sandbox Agents make sense, and how to design safer agent workflows in 2026.