AI Security
Claude code execution and network egress settings explained
What the network options for Claude code execution mean, why all domains is the riskiest, and the mitigations Anthropic lists against prompt injection.
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
Claude Code bash sandbox: what it covers and what it misses
The Claude Code sandbox restricts shell commands only. Defaults for writes, reads and network, how to turn it on, and what runs outside the boundary.
Oct 18, 20263 min read
AI Security
Can an LLM triage security alerts? What it does well and badly
An LLM can summarize, enrich and rank alerts for an analyst, but should not close them alone. See where it helps, where it fails and how to set it up.
Oct 18, 20263 min read
AI Security
Can a system prompt stop prompt injection? Myth versus fact
A system prompt can narrow what a model does, but it cannot enforce a security boundary. Here is what it helps with and what needs code outside the model.
Oct 18, 20263 min read
AI Security
Best of N jailbreaking: why one blocked prompt proves little
Attackers retry the same request with small variations until one gets through. What best of N means for testing, rate limits and how you read your results.
Oct 18, 20263 min read
AI Security
Audit trail for AI analyst actions: what to log
If an AI assistant touches your security tools, you need a record of what it saw, said and did. This guide lists the fields to log and how to keep them.
Oct 18, 20263 min read
AI Security
The 16 MITRE ATLAS tactics explained for LLM applications
A plain-language walk through the 16 MITRE ATLAS tactics, from AI model access to lateral movement, with what each means for an LLM app you run.
Oct 18, 20264 min read
AI Security
ATLAS mitigations checklist for RAG systems
A checklist of MITRE ATLAS mitigations that apply to RAG systems: guardrails, telemetry, dataset provenance, validation, access control and agent limits.
Oct 18, 20264 min read
AI Security
What two ATLAS case studies teach about RAG attacks
Two MITRE ATLAS case studies, the Morris II worm and Slack AI exfiltration, show how a RAG index becomes an attack path. Here are the design lessons.
Oct 18, 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.