AI Security
Should an AI coding agent be allowed to browse the web?
Web access helps agents read documentation but opens paths for prompt injection and data exfiltration. Decide by allowlisting hosts and splitting roles.
Oct 19, 20266 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
Should CI run automatically on a pull request from a cloud agent?
CI runs code with tokens and secrets. Why agent authored pull requests should wait for a human before workflows run, and how to set that up.
Oct 19, 20266 min read
AI Security
Cloud coding agents: how do internet access settings compare?
Documented internet access controls for two cloud coding agents: off by default, allowlists, HTTP method limits, and what to test before you enable more.
Oct 19, 20266 min read
AI Security
What limits stop a scheduled AI task from running away?
Documented caps on routines, loops and schedules, what they protect, and the extra limits you must add yourself to contain a runaway automation.
Oct 19, 20267 min read
AI Security
Do I still need a security scanner if an AI assistant reviews my code?
Yes. An AI reviewer reads and judges code; a scanner looks up known flaws in every library. They do different jobs, and a gate makes sure both happen.
Oct 19, 20268 min read
AI Security
Who should be allowed to promote a model to production?
Fewer people than can train one. A least privilege model for registries, the permission levels to use, and the checks to require before a model goes live.
Oct 18, 20263 min read
AI Security
Which types of tools check the security of an LLM app?
Checking an LLM app takes several tool types: dependency scanners, static analysis, secret scanning, adversarial tests, runtime guardrails and logging.
Oct 18, 20264 min read
AI Security
Which OWASP non-human identity risks apply to AI agents?
OWASP lists ten non-human identity risks. Most map directly to AI agents and MCP servers. See the mapping table and a quick check for each risk.
Oct 18, 20263 min read
AI Security
Which OWASP LLM Top 10 risks matter most for AI coding assistants?
Six of the OWASP Top 10 for LLM Applications apply directly to coding assistants. A mapping of each risk to a developer scenario and a control.
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.