AI Security
What does Allow always mean for an MCP connector tool?
Allow always lets an assistant run a connector tool without asking again. Learn when that is safe for security tools and when to keep the prompt on.
Oct 19, 20266 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 is AI agent identity and which three questions does it answer?
AI agent identity gives each agent its own login so you can answer three questions: who acted, what were they allowed to do, and what did they do.
Oct 19, 20268 min read
AI Security
What data should a read only AI agent be allowed to see?
Read only still exposes data. Scope an AI agent's reads by dataset, row, field and purpose, and know what it can pass on. A decision table and steps.
Oct 19, 20263 min read
AI Security
Does a security vendor's AI see your source code? What to ask
Find out what your security vendor's AI actually receives: code, metadata, or findings only. A short list of data path questions and how to read the answers.
Oct 19, 20263 min read
AI Security
What changed in the 2026-07-28 MCP specification for buyers?
MCP 2026-07-28 is stateless, adds server discovery and deprecates sampling and protocol logging. What that means when you evaluate connectors.
Oct 19, 20267 min read
AI Security
What can go wrong with an AI agent? Five kinds of failure explained
Five ways an AI agent can fail, from honest mistakes to hidden instructions, with a plain example of each and the first control to try for it.
Oct 19, 20268 min read
AI Security
What can an AI desktop app see on your computer? A permission checklist
An AI desktop app may read files, the clipboard, the screen and run commands. A checklist to map each reach, limit it and verify it before rollout.
Oct 19, 20266 min read
AI Security
What can an AI coding assistant see and change on my computer?
A plain guide to what AI coding tools can read, edit and run on your machine, how permission modes and sandboxes limit them, and what to check in settings.
Oct 19, 20269 min read
AI Security
What does an Ollama Modelfile control, and what should you review?
A shared Modelfile sets the weights source, system prompt, example messages, template and parameters. What each instruction does and what to check.
Oct 19, 20264 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.