AI Security
What should nonprofit staff never paste into an AI chat tool?
AI chat tools help a small team write and summarize, but some data should stay out. Here is a plain list of what not to paste and a short rule to share.
Oct 19, 20264 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
What do AI coding security words mean? A plain glossary
Plain definitions of the words you hear about AI coding and software security, from prompt and agent to MCP, SBOM, CVE and gate, with one worked example.
Oct 19, 20269 min read
AI Security
How do AI coding assistants affect SOC 2 dependency controls?
AI assistants add packages quickly, sometimes invented ones. Keep SOC 2 change controls intact with review rules, package checks and a clear policy line.
Oct 19, 20264 min read
AI Security
How do you check an AI-built app before a non-engineer launches it?
A pre-launch check for apps a non-engineer built with an AI assistant: dependencies, secrets, auth, input handling and logs, in plain steps.
Oct 19, 20263 min read
AI Security
What goes in an AI agent tool permission manifest?
A tool permission manifest is a short reviewed file listing what one AI agent may call and with which limits. Here are the fields and a sample layout.
Oct 19, 20263 min read
AI Security
AI agent permissions: five myths and what is actually true
Five common beliefs about AI agent tool permissions, from prompt rules to approval prompts, and what OWASP and MCP texts support. Myth versus fact.
Oct 19, 20263 min read
AI Security
How do you add a new tool to an AI agent with deny by default?
Deny by default means a new tool is blocked until reviewed. Seven steps to request, review, test, approve and monitor a new tool for an AI agent.
Oct 19, 20263 min read
AI Security
What happens when an AI agent runs a task, and where do controls fit?
Follow one invented AI agent task from request to record, and see where identity, limits, approval, and logging fit at each of its seven steps.
Oct 19, 20269 min read
AI Security
Is a dataset just data, so it is safe? Myth versus fact
Datasets can carry poisoned examples, loading code, restrictive licenses and sensitive records. Five myths about training data checked against sources.
Oct 19, 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.