AI Security
LLM06 excessive agency: which controls work and which do not?
OWASP LLM06 excessive agency comes from too many functions, permissions and autonomy. Controls that prevent it, and controls that only limit the damage.
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
OWASP LLM06 excessive agency: a checklist for engineering leads
Excessive agency means an AI agent can do more than it should. Use this checklist to cut tools, narrow permissions, add approvals and check downstream.
Oct 18, 20264 min read
AI Security
LLM05 improper output handling: which controls work and which do not?
Treat model output as untrusted input. Encode it for each destination, parameterize queries and limit privileges. Prompt wording alone is not a control.
Oct 18, 20264 min read
AI Security
LLM05 improper output handling: which compliance controls does it touch?
Improper output handling is classic injection with a model in the middle. See how OWASP LLM05 maps to secure development, logging and validation duties.
Oct 18, 20265 min read
AI Security
How do you test for OWASP LLM04 data and model poisoning?
Test for LLM04 by checking where data and model files came from, comparing behavior against a fixed evaluation set, and watching training for anomalies.
Oct 18, 20265 min read
AI Security
OWASP LLM04 data and model poisoning: what it is and how to defend
LLM04 covers tampered training, fine-tuning and embedding data, plus poisoned model files. Learn the documented examples and the controls OWASP recommends.
Oct 18, 20264 min read
AI Security
OWASP LLM03 supply chain: what to log and alert on
OWASP LLM03:2025 covers models, adapters, datasets and packages you did not build. Here is a concrete list of events to log and the alerts worth wiring up.
Oct 18, 20264 min read
AI Security
What is LLM03 supply chain risk in the OWASP Top 10 for LLM apps?
LLM03 covers risk from packages, models, adapters and data you did not build. See the incidents OWASP cites and the controls that cut the exposure.
Oct 18, 20264 min read
AI Security
OWASP LLM03 supply chain: how it maps to compliance duties
LLM03 supply chain risk lines up with NIST SP 800-161, the NIST AI RMF and EU AI Act Articles 17 and 25. See the mapping and the evidence to keep for each.
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.