AI Security
How do you cap resource abuse on a self-hosted inference server?
Cap request size, output length, parallel samples, queue depth and per-user rate. A practical checklist built on OWASP LLM10 and vLLM's own limits.
Oct 19, 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 limit AI agent tool arguments: paths, recipients, amounts?
Allowing a tool is not enough. Constrain its arguments too: file paths, email recipients, record IDs and amounts, checked by code outside the model.
Oct 19, 20263 min read
AI Security
Can AI agents get tool access just in time instead of standing access?
Yes. Grant an AI agent a risky tool for one task or a short window, then let it expire. How time boxed tool grants work and which tools deserve them.
Oct 19, 20263 min read
AI Security
What isolation should a coding agent have if it skips prompts?
Skipping permission prompts removes the human check. Anthropic says to run such sessions in a container, VM, or sandbox runtime. Here is what to set up.
Oct 19, 20266 min read
AI Security
Is torch.load safe? What weights_only does and does not cover
torch.load unpickles by default, so the file source matters. See what weights_only=True restricts, what it leaves alone, and how to load checkpoints.
Oct 19, 20264 min read
AI Security
Is torch.hub.load safe? What hubconf.py runs on your machine
torch.hub.load fetches a GitHub repo and runs Python from its hubconf.py. What trust_repo does, the defaults to know, and how to use it with care.
Oct 19, 20265 min read
AI Security
Is loading from TensorFlow or Flax checkpoints a way around pickle?
Hugging Face notes TF and Flax checkpoints are not affected by pickle attacks and can be converted on load. When that helps, and where it stops.
Oct 19, 20263 min read
AI Security
Is model output safe to pass to your database, shell or browser?
Treat LLM output as untrusted input. OWASP's improper output handling guidance: validate, encode and parameterize before output reaches another system.
Oct 19, 20263 min read
AI Security
Is it safe to connect an AI assistant to your email, calendar and files?
It can be safe if you limit what the assistant can read and do. Use this plain checklist before you connect an AI assistant to email, calendar, or files.
Oct 19, 20268 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.