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.
Map tools to the OWASP LLM Top 10 for model-centred apps and to the Agentic Top 10 when software plans and acts. Many teams need both, so check design.
The ten OWASP LLM Top 10 2025 categories, and six questions that turn a vague coverage claim into something you can verify before you buy it.
A decision guide to the privacy controls in consumer AI chat tools, from Claude and Gemini docs: training opt-out, delete, incognito and unsharing.
SP 800-218A is an SSDF community profile for generative AI and dual-use foundation models. It adds to the SSDF, it does not replace it. Who it is for.
NIST's CAISI tested agent hijacking and found new attacks far outperformed known ones, and that retries raised success. What to change in your own tests.
The UK NCSC says prompt injection may never be fully fixed. What that means for how you design agents, and the four actions it recommends you take.
The NCSC secure AI development guidelines have four stages. Map each stage to a control you can check in a product, from threat modelling to monitoring.
Keep a frozen test set, rerun it after each model or setting change, compare with the last accepted results, and ask for notice and pinning in contracts.
Which model files can run code when loaded? A quick table for .bin, .pt, .h5, .keras, SavedModel, .onnx and .safetensors, with sources and review advice.
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.
