AI Security
How do you answer 'Do AI agents have their own identity?'
Reviewers ask whether agent actions can be told apart from human ones. Answer with identity type, credential owner, scope and revocation in plain terms.
Oct 19, 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
How should security tooling treat prompt and agent skill files?
Prompt and skill files steer agents and live beside code. Treat them as reviewed, versioned, owned artifacts, and ask your tooling whether it sees them.
Oct 19, 20263 min read
AI Security
How do secrets end up in an AI coding agent's context?
Secrets reach a coding agent through project files, terminal output, pasted logs and tool results. Here are the five paths and a fix for each.
Oct 19, 20263 min read
AI Security
How can prompt injection reach a coding agent through your repository?
Prompt injection reaches a coding agent through issues, pull requests, docs, dependencies and tool results. Here is the input map and a control for each.
Oct 19, 20263 min read
AI Security
How does OAuth sign-in work when you connect a security connector?
Connecting a remote MCP connector means finding the login server, approving scopes and getting a token. The steps in plain words, with what to check.
Oct 19, 20266 min read
AI Security
How much access should an AI coding agent have to your systems?
Give a coding agent the fewest tools and permissions its task needs, with human approval for high-impact actions. A guide built on OWASP excessive agency.
Oct 19, 20263 min read
AI Security
How is an MCP server different from an IDE extension for risk?
Both run code with the user's privileges. The trigger differs: an extension runs its own logic, while a model can call an MCP tool after reading text.
Oct 19, 20264 min read
AI Security
How does an AI agent use a tool through MCP, step by step?
Follow one tool call from your request to the answer: discovery, choice, the call, the result, and where a person or a guardrail can step in.
Oct 19, 20268 min read
AI Security
How does a golden path help with AI-generated code?
AI speeds up coding, but gains can be lost downstream. How a golden path tests, scans and gates AI-written changes the same way as any other change.
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.