AI Security
Agent output handling: treat the model like any other user
Model output can carry XSS, SQL and shell payloads into your backend. OWASP's zero-trust advice: validate, encode and parameterize before anything runs.
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
Agent delegation vs impersonation: what RFC 8693 distinguishes
When an AI agent acts for a user, should it impersonate them or act as a delegate? RFC 8693 defines both and the actor claim that records who did what.
Oct 18, 20263 min read
AI Security
What to log for every AI agent tool call
A field list for agent tool call audit logs, drawn from MCP guidance on logging, delegation and elevation, so incidents can be reconstructed afterward.
Oct 18, 20263 min read
AI Security
Which OWASP LLM risk to tackle first: chatbot, RAG app or agent
A decision guide to the ten OWASP LLM risks by app type: where to start for a plain chatbot, a RAG assistant and a tool-using agent, with reasons.
Oct 18, 20263 min read
AI Security
Who should red team your AI system? NIST's four team types
NIST's Generative AI Profile describes general public, expert, combination and human plus AI red teams. How to pick the right mix for your system.
Oct 18, 20263 min read
AI Security
Who should sit on an AI governance committee?
An AI governance committee needs the right seats: security, legal, privacy, data, engineering and a business owner. Charter, cadence and decision rights.
Oct 18, 20263 min read
AI Security
Why AI agent bills balloon: the hidden costs inside agent loops
Agents resend context on every step, run idle check-ins and spawn helpers. Here are the documented sources of token growth and how to cap each one.
Oct 18, 20263 min read
AI Security
Will AI cause mass cyberattacks by 2027? What the evidence says
The 2026 International AI Safety Report says AI helps attackers at many steps, but fully autonomous end-to-end attacks have not been reported.
Oct 18, 20263 min read
AI Security
XML tags in security prompts: structure, not a security boundary
Tags help a model separate instructions from data but do not stop prompt injection. What they give you, and what still needs checks outside the model.
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.