AI Security
Is loading a Keras model file safe? What safe_mode blocks
Keras load_model keeps safe_mode on by default for the v3 format, which blocks unsafe lambda deserialization. What it covers, and when to turn it off.
Oct 18, 20263 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
Is it safe to paste code or secrets into an LLM chat?
Never paste live secrets into a chat. Code is a policy question. Here is what to redact, which settings to check, and what to do after a slip.
Oct 18, 20263 min read
AI Security
Is it safe to load a scikit-learn joblib or pickle model file?
Only load a joblib or pickle model from a source you trust. scikit-learn's own docs say so. Here are safer formats, version rules and a quick checklist.
Oct 18, 20263 min read
AI Security
Is it safe to install an MCP server from the internet?
A local MCP server is code that runs with your privileges. Read the launch command, check the source, sandbox it, scope its access and review its tools.
Oct 18, 20264 min read
AI Security
Is it safe to expose a Ray cluster dashboard to the network?
No. Ray's security docs say anyone who can reach the dashboard, jobs or client ports can run arbitrary code. How to isolate the cluster and add token auth.
Oct 18, 20263 min read
AI Security
Is an inference server CVE exploitable in my deployment? Five questions
Five questions that decide whether a serving runtime CVE can hurt you: version, feature in use, network reach, authentication, and what an attacker gains.
Oct 18, 20263 min read
AI Security
Is a third-party MCP server safe to connect to your AI tools?
Treat a third-party MCP server like a vendor with code execution: check local versus remote, consent prompts, token handling and scopes before you connect.
Oct 18, 20263 min read
AI Security
Is a TensorFlow SavedModel safe to load? Models are code
TensorFlow says untrusted models are equivalent to untrusted code. Here is what that means for SavedModel files, checkpoints, and how to load them.
Oct 18, 20263 min read
AI Security
Is a model-serving endpoint different from a normal API for security?
A model endpoint is still an API, so the usual API risks apply, but it adds unauthenticated control routes, cost abuse, and tenant leakage to plan for.
Oct 18, 20263 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.