AI Security
Match an LLM's privileges to whoever wrote its input
A design rule for LLM security automation: when the model reads content from a party, give it no more power than that party has. Examples and trade-offs.
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
Why wrong LLM answers are a security issue: OWASP LLM09
Credible but false output can cause breaches, legal exposure and bad code. What OWASP says about LLM misinformation and the controls that reduce it.
Oct 18, 20263 min read
AI Security
LLM API settings that help prevent model extraction and abuse
If you expose your own model through an API, limit logprobs, input size, rate and access. The OWASP controls against extraction, replication and abuse.
Oct 18, 20263 min read
AI Security
Least privilege for AI coding agents: a permissions checklist
What access a coding agent actually needs, what to withhold, and how to scope files, shell, network and credentials so one bad instruction stays small.
Oct 18, 20263 min read
AI Security
Least privilege for CI review bots and coding agents
What permissions an AI review bot or coding agent needs in CI, what to withhold, and how to check the token it runs with. A practical checklist.
Oct 18, 20263 min read
AI Security
Keras safe_mode and Lambda layers: why model files can run code
Keras load_model has a safe_mode flag that blocks unsafe Lambda deserialization in v3 files. Here is what it covers and why you should leave it on.
Oct 18, 20264 min read
AI Security
Jailbreak or prompt injection: which term fits your incident?
Jailbreaking is a form of prompt injection aimed at removing safety rules. A short guide to choosing the right term in tickets, reports and audits.
Oct 18, 20263 min read
AI Security
Is the NIST AI RMF mandatory? What voluntary really means
The NIST AI RMF is voluntary and not a certification, but contracts, customers or other rules may still point to it. How closely should you follow it?
Oct 18, 20263 min read
AI Security
Is an LLM judge guardrail enough to stop prompt injection?
A second model that screens inputs and outputs is useful, but it can be injected too. Why guardrail models are one layer and what to pair them with.
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.