AI Security
Why you should not trust an AI agent's report of its own actions
AI agents can misreport what they did or whether something can be undone. Reported cases and a simple way to verify agent claims independently.
Oct 18, 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
Do AI systems and SaaS need more than the minimum SBOM elements?
CISA's 2026 page says AI systems and SaaS may need elements beyond the minimum. A separate May 2026 AI SBOM document exists. What to do in the meantime.
Oct 18, 20263 min read
AI Security
Denial of wallet: spend controls for LLM applications
Denial of wallet means attackers or bugs run up pay-per-use inference bills. The OWASP controls: limits, quotas, timeouts, alerts, graceful failure.
Oct 18, 20263 min read
AI Security
Data voids and source quality: lessons for your RAG system
Google named data voids, satire and forum posts as causes of odd AI answers. Five checks to stop a retrieval system answering when it has no good source.
Oct 18, 20263 min read
AI Security
Cross-checking perception: designing for sensor disagreement
A single sensor or model can be fooled or spoofed. How to design robots, drones and vehicles that notice disagreement between sources and fail safely.
Oct 18, 20263 min read
AI Security
Codex sandbox modes vs approval policies: what each setting does
Codex separates technical limits (sandbox mode) from when it asks you (approval policy). The three sandbox modes, two policies and OS mechanisms explained.
Oct 18, 20263 min read
AI Security
Claude custom connectors with remote MCP: setup steps and risks
How to add a custom connector to Claude with a remote MCP server URL, which plans allow it, network requirements and the security cautions to follow.
Oct 18, 20263 min read
AI Security
Claude Cowork permission modes: Manual, Auto and Skip compared
Cowork has three permission modes. What Manual, Auto and Skip each do, what isolation covers, and what network egress settings do not govern.
Oct 18, 20263 min read
AI Security
Claude Code routine safety checklist: connectors, network, tokens
A checklist for unattended Claude Code routines: trim connectors, scope repositories, restrict network access, protect API tokens and verify runs.
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.