AI agents.

Limit agent permissions, tool access, approval scopes and action budgets.

Agentic AI Prompt Injection: How Assistants Get Hijacked—and How to Contain Them

A practical 2026 lab design for defending AI assistants: isolate untrusted content, constrain tools, bind approvals to actions and measure policy failures.

9 min read Oct 21, 2026

Artificial intelligenceDeep Dive

AI Agent Actions: Budgets, Approvals and a Working Stop Button

Constrain an AI agent’s actions with server-side limits, parameter-bound approvals, idempotent execution and cancellation tests for queued work.

4 min read Oct 28, 2026

Artificial intelligenceDeep Dive

MCP Authorization in 2026: Tokens, Audiences and Trust Boundaries

Review HTTP-based MCP authorization against the July 2026 specification, with audience checks, separate downstream grants and safe metadata handling.

4 min read Oct 28, 2026