AI Security
Is asking a second LLM to check the first one real verification?
A second model can catch some mistakes, but it is not independent proof. Learn when cross-checking helps and which checks need real evidence.
Oct 19, 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 a PyTorch model a program? What the PyTorch security policy says
Yes. PyTorch's security policy says models are programs, so an untrusted one is like untrusted code. What to do about formats and sandboxes.
Oct 19, 20264 min read
AI Security
Is a prompt from a public library safe to paste into production?
Not without review. Copied prompts can carry hidden text, unwanted behavior and unclear rights. A short vetting routine before adopting one.
Oct 19, 20263 min read
AI Security
Is a custom AI connector verified by the chat provider?
No. Both Anthropic and OpenAI say custom connectors are not verified by them. What that means for buyers, and the checks you must run yourself.
Oct 19, 20266 min read
AI Security
Inventory first or enforcement first when securing AI agents?
Start with inventory unless you already know your agents. A decision guide for choosing between visibility and enforcement when buying AI agent security.
Oct 19, 20264 min read
AI Security
ClamAV versus a pickle import scan: what does each catch in a model?
An antivirus scan and a pickle import scan answer different questions about a model file. What each checks, what each misses, and what to run yourself.
Oct 19, 20263 min read
AI Security
How do you vet an MCP server before a coding agent connects to it?
Vet an MCP server by checking its source, startup command, scopes, and network use before connecting. A seven step checklist for security-minded teams.
Oct 19, 20267 min read
AI Security
How do I triage AI-generated security pull requests as a maintainer?
A short triage routine for AI-written security pull requests: demand a reproduction, a failing test, and a small diff before you spend review time.
Oct 19, 20264 min read
AI Security
How do you threat model an AI coding agent in one hour?
Threat model an AI coding agent in one hour: list what it can reach, what it reads, who can steer its input, and which actions need a human yes.
Oct 19, 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.