AI Security
Safeguard articles tagged "AI Security" — guides, analysis, and best practices for software supply chain and application security.
100 articles
Narrow tools or a general shell for AI agents: which is safer?
Narrow, purpose built tools are easier to scope, approve and audit than a general shell or fetch tool. A comparison and a path to replace open ended tools.
MCP versus a REST API for security tools: what is the difference?
A REST API is built for programs written by developers, while MCP lets an AI assistant discover and call tools itself. A comparison for security buyers.
How do MCP tool approvals work when an assistant task runs unattended?
Unattended tasks cannot click Allow. How tool approvals behave in scheduled runs, which tools still stall, and how to add a guardrail layer.
Should MCP tool allowlists differ for dev, staging and production?
Yes. Use a looser MCP tool allowlist in dev and a tight one in production, with promotion steps between. A table of what to allow where, with checks.
Do MCP servers in a cloud agent sandbox bypass your egress rules?
Sandbox firewalls may not cover MCP server processes or connector traffic. How to find that gap and close it for cloud coding agents that use MCP tools.
What MCP authorization checks should a buyer expect from security tooling?
Buyers should expect security tooling to help spot MCP servers that skip token audience checks, pass tokens through, or put tokens in URLs. What to ask.
What do MCP and AI agent terms mean? A plain English glossary of 20
Twenty MCP and AI agent terms defined in plain English, from agent and tool call to prompt injection and guardrail, grouped so you can learn them in order.
Are these MCP and AI agent security myths true? Eight myths vs facts
Eight common beliefs about MCP and AI agent security, checked against OWASP, NIST and the MCP specification, with what to do about each one.
How do you map the OWASP LLM Top 10 to questionnaire answers?
Use the OWASP LLM Top 10 as a checklist for AI assistant and MCP answers: for each risk, one sentence on exposure, one control, one piece of evidence.
Local or remote MCP for security tools: which should you choose?
Local MCP servers run on a laptop with your privileges, remote ones run on the vendor side. A decision guide for security tools, with a comparison table.
Local or hosted AI coding agent: where does your code actually go?
Local agents still send context to a model service, and hosted agents run your code in someone else's environment. Know both data paths before you choose.
How do you cap resource abuse on a self-hosted inference server?
Cap request size, output length, parallel samples, queue depth and per-user rate. A practical checklist built on OWASP LLM10 and vLLM's own limits.
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.