AI Security
Agent credentials: issue, scope, expire and revoke
A lifecycle for the credentials an AI agent holds: one identity per agent, narrow scopes, short lifetimes, monitored use and a revoke path you have tested.
Oct 18, 20263 min read
Resources
Supply Chain Security, in plain English.
Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
Filter
Articles
RSS feedAI Security
Which AI tools belong in your AI inventory? A decision guide
Does an AI feature inside a vendor app belong in your AI inventory? A decision guide for in-house models, APIs, embedded features and agents.
Oct 18, 20263 min read
AI Security
Which AI agent tool calls need human approval?
A decision guide for human-in-the-loop approval on agent tool calls, based on the MCP spec and OWASP guidance: irreversible, external and costly actions.
Oct 18, 20263 min read
AI Security
Which AI agent actions should ask for approval first?
A simple way to sort agent tool calls into allow, ask and block, using impact, reversibility and reach, with OWASP and MCP guidance behind it.
Oct 18, 20264 min read
AI Security
When to pause, roll back or retire an AI system
Criteria for stopping an AI system: unacceptable risk, drift, vendor change or lost purpose. What the NIST AI RMF says about disengaging and retiring.
Oct 18, 20263 min read
AI Security
When not to use AI in security operations
AI is a poor fit for some SOC tasks. Here are the cases to avoid or keep manual: irreversible actions, legal calls, unmeasured alerts and sensitive data.
Oct 18, 20263 min read
AI Security
What to put in an AI acceptable use policy
An AI acceptable use policy should cover scope, approved tools, data rules, human review, incidents and exceptions. A section by section outline to adapt.
Oct 18, 20263 min read
AI Security
What to log in a RAG system so you can investigate incidents
The fields a RAG system should log, from retrieved chunk IDs to tool calls and agent identity, so you can trace a bad answer back to its source.
Oct 18, 20263 min read
AI Security
What should never go in an LLM system prompt: a checklist
Assume users can read your system prompt. A checklist of what to keep out, such as credentials, permission maps and business limits, and where to put it.
Oct 18, 20263 min read
Stay informed
Weekly insights on software supply chain security, delivered to your inbox.
Self-healing security runs on Safeguard.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.