CodeCrux / Field notes
163 insights / and countingPractical thinking on the systems, controls and trade-offs that make AI useful in production.
AIML
Learn how to contain browser-using AI agents with per-task profiles, delegated short-lived credentials, and policy checks that treat page content as hostile input.
Read the latest insight
A comprehensive tutorial covering implement oauth 2.1 authorization in mcp servers: a hands-on guide to anthropic's agent security s...
Read insight
Stop system prompt leakage: learn practical AI agent security threats detection with canary tokens, prompt watermarks, and repeatabl...
Read insight
Learn AI agent security best practices by building a redaction and egress proxy that inspects tool calls, blocks data exfiltration, ...
Read insight
Learn how to build hash-chained audit logs that give your AI agent security audit tamper-evident traces for forensics, governance, a...
Read insight
Learn how ai agent security tools verify third-party MCP servers and plugins with manifest signing, permission diffing, and version ...
Read insight
Learn how to contain AI agent security risks with hard budgets, rate limits, loop circuit breakers, and a policy boundary that keeps...
Read insight
Learn how to contain AI agent security threats by sanitizing untrusted tool output, enforcing policy, and building a fail-closed tru...
Read insight
Learn how to protect persistent AI agent memory from poisoning with signed writes, provenance checks, scoped retrieval, and fail-clo...
Read insight
A comprehensive tutorial covering delegated identity for ai agents: implement short-lived capability tokens with step-up auth. Pract...
Read insight
A comprehensive tutorial covering hunting multi-step tool attacks: reproduce and mitigate chained agent exploits. Practical examples...
Read insight
Learn how to trace, baseline, and alert on AI agent tool abuse using a practical AI agent security monitoring meta practical ai agen...
Read insight
Learn how to build an MCP policy gateway to enforce zero-trust controls, limit tool access, and create auditable trails for AI agent...
Read insight
A comprehensive tutorial covering secure agent identity with sailpoint and silverfort: entitlement reviews and step-up auth for tool...
Read insight
Learn how to detect, constrain, and monitor autonomous AI agents using Wiz, Zenity, and CrowdStrike with a practical, policy-first l...
Read insight
A comprehensive tutorial covering deploy ai agent security infrastructure on kubernetes: policy gateway, mtls sidecars, and audit si...
Read insight
Learn how to replay multi-step tool attack data from Kaggle as a labeled regression suite that catches agent tool abuse before produ...
Read insight
Learn the five key principles of AI agent security and implement each control in a reference MCP server, with policy checks, approva...
Read insight
Learn how to sandbox code-executing AI agents with container isolation, seccomp profiles, and egress allowlists, plus policy, identi...
Read insightHave a production AI problem?
Talk to an AI architect about the workflow, the constraints and the path from prototype to production.
