AI Security
How to test an LLM app for misinformation (OWASP LLM09)
OWASP LLM09 covers confident but false model output. Build a golden question set, check grounding, verify suggested packages and keep a human in the loop.
Oct 18, 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
How do you test an LLM app for improper output handling?
Test LLM05 by listing every place model output lands, pushing harmless marker strings through each, and checking it is encoded or validated. Steps inside.
Oct 18, 20265 min read
AI Security
How do you test an app for OWASP LLM07 system prompt leakage?
Test for system prompt leakage by assuming the prompt will be read, then hunt for secrets and rules in it and check controls outside the model. Defensive.
Oct 18, 20264 min read
AI Security
How to test your LLM app for excessive agency (OWASP LLM06)
Excessive agency means your LLM can do more than it should. Here is a defensive test plan built on OWASP LLM06: tools, permissions, approvals and logs.
Oct 18, 20264 min read
AI Security
AI agents and API keys: why long lived keys are the wrong default
Give each AI agent its own identity and short lived credentials instead of a shared API key. Here is how to scope, expire, log and revoke agent access.
Oct 18, 20265 min read
AI Security
Small models as security classifiers: mistakes teams make
Small models can screen prompts and outputs cheaply, but teams treat them as walls. Six common mistakes and how to test a classifier honestly.
Oct 18, 20264 min read
AI Security
Small models as security classifiers: a hands-on checklist
A small model can screen prompts and outputs cheaply, if you test it on your own data. Use this checklist for labels, thresholds, failure modes and drift.
Oct 18, 20264 min read
AI Security
How do you block hallucinated packages (slopsquatting)?
AI coding tools sometimes suggest packages that do not exist, and attackers can register them. Block it with allow lists, history checks and review.
Oct 18, 20265 min read
AI Security
Self-hosted open models: which security work they handle well
A self-hosted open model has no plan limit, only your hardware. Which security tasks fit, which do not, and what to check first. Prices read 18 Oct 2026.
Oct 18, 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.