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.
Log prompts, retrieved sources and filter hits for LLM02, then alert on sensitive patterns in outputs, odd retrieval and repeat probing. A defensive guide.
LLM02 is when a model or its app reveals PII, secrets or business data. See OWASP's scenarios, a real world reference and the defensive controls to apply.
LLMs can reveal PII, credentials and business data through outputs and apps around them. Use OWASP's controls: sanitize, limit access, redact, set policy.
OWASP's prompt injection page names no compliance frameworks. See how it lines up with NIST AI 600-1, EU AI Act Article 15 and MITRE ATLAS, and the limits.
OWASP says no prevention is fool-proof, so limit impact: constrain the model, validate output, least privilege, human approval, testing. A lead checklist.
An agent that can act on equipment is a new actor with credentials and tools. Treat its output as untrusted, limit permissions, keep humans on risky steps.
Map the NIST AI RMF functions Govern, Map, Measure and Manage to supply chain controls: inventories, provenance, scanning and response. A working map.
Loading a pickle model file can run code, and published weights have weak provenance. Learn five common mistakes and the safer habits that replace them.
A model card should state intended use, limits, evaluation, data sources and a security contact. These are the mistakes teams make when they publish one.
Weekly insights on software supply chain security, delivered to your inbox.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.
