Blog
Security articles
Every article we have written on security, newest first. Written by the engineers who build these systems, for the people deciding what to build.
10 articles · page 1 of 2
SecurityAn AI Agent Deleted a Production Database in 9 Seconds. Here's the Architecture That Makes It Impossible
An AI agent deletes production data when it holds a credential that allows it. Scoped tokens, a policy gate, bound approvals and off-site backups stop it.
30 Oct 202610 min
SecurityAI Finds Vulnerabilities Faster Than Teams Patch: A 2026 Security Checklist for Teams Shipping AI
AI now finds and exploits flaws faster than most teams patch. A checklist from October 2026: same-day patching, SBOMs, fail-closed scans and keys you can rotate.
30 Oct 202611 min
SecurityAfter the Railway and OpenCode Leaks: Keeping API Keys and Secrets Away From AI Agents
Secrets in env vars and files are within reach of every process and agent. Give agents short-lived, scoped tokens via a broker, and rotate without redeploying.
30 Oct 202611 min
SecurityYour AI Agent Has Your Admin Token: Least-Privilege Access with OAuth Token Exchange and Entra Agent ID
Give each AI agent its own identity and short-lived, single-API tokens. When it acts for a person, use OAuth token exchange so it never exceeds their rights
30 Oct 202610 min
SecurityLLM Guardrails Compared for 2026: Bedrock Guardrails vs Azure Prompt Shields vs NeMo vs Guardrails AI
Bedrock Guardrails and Azure Content Safety are managed text checks; NeMo and Guardrails AI run in your code. What each catches, and what none of them do.
30 Oct 202611 min
SecurityMCP Security in 2026: The STDIO Flaw, 200,000 Exposed Servers, and a Checklist Before You Connect One
MCP is safe only if no untrusted party can write a server config. What the April 2026 STDIO disclosure means, and eight checks before you connect a server.
30 Oct 202611 min
SecurityPrompt Injection in AI Agents: Real Attacks, and Why Filters Alone Won't Stop Them
Prompt injection hijacks an AI agent through text it reads. Real attacks, why filters only reduce the risk, and the structural fixes that limit the damage.
30 Oct 202611 min
SecurityHow to Red-Team an AI Agent Before Launch with Promptfoo, PyRIT and garak
Red-team an AI agent by attacking its tools, not just its words: Promptfoo for app tests in CI, PyRIT for multi-turn attacks, garak for broad model probes.
30 Oct 202610 min
SecuritySystem Prompts Are Not Security Controls: Where AI Agent Guardrails Actually Belong
A system prompt steers an AI agent, enforces nothing. Where each guardrail belongs, from filters to policy checks to database constraints, and what each stops.
30 Oct 202611 min
Tell us which process is costing you.
Thirty minutes, no preparation, no deck. You describe what keeps going wrong and we tell you whether AI is the answer — including when it is not.
Prefer email? contact@devs-core.com