15 Checks Non-Developers Should Keep in Mind Before Shipping AI-Written Code to Production

AI coding tools produce working code at astonishing speed. Working, however, is not the same as production-grade. Here are fifteen concrete things to look at before you ship what your AI wrote.

Letting AI Agents Use the AWS CLI Safely — A 2026 Defense-in-Depth Design That Disarms Any Single Credential

Hand an AI agent a shell and AWS credentials at the same time, and a single prompt injection can translate straight into deleted cloud resources or exfiltrated secrets. IAM Identity Center plus ReadOnlyAccess is the right starting direction. But ReadOnlyAccess is not 'read-only equals safe,' and tool-side deny rules do not survive adversarial evasion. Working from the premise that the only trustworthy enforcement boundaries are IAM and OS sandboxes, we layer defenses across authentication, authorization, execution, approval, and audit so that seizing a repository or a shell alone never reaches production.

Letting AI Agents Use the Azure CLI Safely — A Defense-in-Depth Design with Managed Identity, a Command Broker, and Resource Locks

Hand an AI agent a shell and Azure credentials at the same time, and a single prompt injection can turn into a stray az command that deletes resources or exfiltrates secrets. Short-lived credentials and least privilege, as Microsoft recommends, is the right direction. But a service principal's client secret is highly persistent once leaked, and tool-side deny rules do not survive adversarial evasion. Working from the premise that the only trustworthy enforcement boundaries are RBAC, organization policy, and OS sandboxes, we layer defenses across authentication, authorization, execution, approval, and audit so that seizing a single input never reaches production.

Replacing Checkmarx with Open Source — Designing an AppSec Stack from SAST, SCA, IaC, and Secrets

No single open-source tool replaces Checkmarx one to one. The realistic answer is a loosely coupled multi-engine setup: place a dedicated OSS scanner in each functional category and bind them with a management layer like DefectDojo. This article walks through the Semgrep CE versus Opengrep split in SAST, Trivy-centered SCA, Checkov and KICS for IaC, Gitleaks and TruffleHog for secrets, and DefectDojo's deduplication and Jira sync, as of 2026, along with the limits such as legacy languages and supply-chain risk.

Detecting Terraform Vulnerabilities with Checkov: An Insecure vs Secure Comparison Demo

IaC security should be verified before you apply. Checkov lets you detect Terraform vulnerabilities through static analysis. Using five AWS resources (S3, Security Group, RDS, IAM, CloudTrail), this demo reproduces the journey from 37 FAILEDs to 111 PASSEDs in a public repository.

How to Get Audit Logs on Claude Team Plan Without Upgrading to Enterprise

Claude Team plan lacks audit log functionality. Rather than upgrading to Enterprise, you can build equivalent visibility by combining API gateways, browser monitoring, and CASB/DLP controls. This post covers specific tool choices and configuration examples for each layer.

How to Get Audit Logs on Cursor Teams Without Upgrading to Enterprise

Cursor Teams plan lacks audit log functionality, and even Enterprise audit logs only record management events, not prompt content or generated code. This post covers how to build audit trails using Cursor Hooks as the primary mechanism, supplemented by API gateways and SSO logs.

Comparing Open-Source DAST Tools — Binding ZAP, Nuclei, Wapiti, and Nikto into a DevSecOps Pipeline

DAST (dynamic application security testing) hits a running app from the outside with simulated attacks to surface runtime-specific vulnerabilities that SAST and SCA miss. Open-source DAST has moved from 'one tool does everything' to 'divide the work among lightweight tools.' Centered on the all-around ZAP, template-driven Nuclei, lightweight-CLI Wapiti, and server-exposure-focused Nikto, this article adds the SQLi specialist sqlmap and the XSS specialist Dalfox, plus the free-but-not-OSS Dastardly and StackHawk, and lays out 2026 selection criteria alongside SPA/API coverage, phased CI/CD integration, and false-positive tuning.

How to Disable Password Manager Auto-save and Auto-fill for input type="password"

Password managers like 1Password and LastPass are very convenient, but there are situations where developers want to control auto-fill and save suggestions. This article explains HTML attributes to disable those features for specific input fields.

Letting AI Agents Use the Google Cloud CLI Safely — Drop the Keys, Lock It Down with Impersonation and a Command Broker

Hand an AI agent a shell and Google Cloud credentials at the same time, and a single prompt injection can turn into a misfired gcloud command that escalates privileges or exfiltrates data. Short-lived credentials and least privilege, which is Google's official guidance, are the right direction. But service account keys persist dangerously when leaked, and tool-side deny rules do not survive adversarial evasion. Working from the premise that the only trustworthy enforcement boundaries are IAM, organization policy, and OS sandboxes, we layer defenses across authentication, authorization, execution, approval, and audit so that seizing a single input never reaches production.

Why the Gemini CLI Security Extension Impressed Me

I tried the Gemini CLI Security extension and it turned out to be impressively capable. While I've been using Claude Code and Codex with their defaults, I now want to actively leverage Gemini CLI extensions as well.

gh-security-scan: An AI Coding Assistant Skill for Cross-Organization Supply Chain Attack Investigation

Supply chain attacks like the axios npm account takeover are becoming more frequent. Investigating their impact across an entire organization is time-consuming manual work. We built gh-security-scan, an AI coding assistant skill that automates cross-organization security investigation. It supports Claude Code, Codex, and Gemini CLI.

Granting Google Cloud Access by Team — Access Groups and Folders as Two Axes

"Folders or group addresses?" is a false choice. Folders scope permissions, groups identify principals, and the right design uses both axes together.

Library Vulnerability Update Policy Guide - Designing Severity-Based SLAs from Global Standards

'How many days do we have to patch a Critical vulnerability?' — There is no single authoritative answer. This article surveys major global frameworks, extracts the few that prescribe concrete timelines, and synthesizes them into a unified SLA design that is audit-ready and operationally practical.

Bug Bounty Programs in Open Source Projects - Lessons from Implementation and Termination Cases

Investigating bug bounty program implementation and termination cases in open source projects. From big tech-led programs to OSS companies like Vercel, Supabase, and Neon, and new challenges in the AI era, we explore the current state of OSS security.

Introducing SECURITY.md and security.txt to Open Source Projects - Learning from Giselle

Learn how to introduce SECURITY.md and security.txt to open source projects using Giselle as an example. Covers security policy writing, vulnerability reporting methods, and automating update reminders with GitHub Actions.

OWASP Security Standards Adopted by Major Platforms - Learning from Google, Microsoft, Salesforce, and Slack

We introduce cases where major platforms have adopted OWASP as a security review standard for web application development. The fact that giant platforms like Google, Microsoft, Salesforce, and Slack have adopted OWASP as their standard demonstrates that it serves as the security baseline for modern web application development.

How pnpm's minimumReleaseAgeExclude Helped with Emergency Security Updates

pnpm's minimumReleaseAge is a useful setting that delays installing freshly released packages to improve stability. However, when you need to urgently apply security patches, this setting can work against you. This article introduces a real-world example of using minimumReleaseAgeExclude to exclude specific packages from the restriction, enabling rapid and safe security update responses.

Free Tools to Check SPF, DKIM, and DMARC Settings

SPF, DKIM, and DMARC settings are essential for email security and deliverability. This article introduces free tools to check these settings and compares their features. Use these tools to enhance your email reliability.

Supabase Privacy and Security

Detecting Terraform Misconfigurations with Trivy: A Comparison Demo with Checkov

Following our Checkov demo, we scanned the same Terraform code with Trivy. Checkov found 37 issues; Trivy found 27. Where does the gap come from? We compare severity classification, IAM wildcard handling, and check ID systems using actual scan results.

Beyond Trivy and tfsec - Comparing Alternative Security Scanning Tools (Grype, Checkov, Snyk, KICS)

The Trivy supply chain compromise and tfsec's maintenance-mode status have made it urgent to evaluate alternative security scanning tools. This article compares options beyond Trivy and tfsec: Grype, Snyk, and Docker Scout for vulnerability scanning, and Checkov and KICS for IaC scanning, with practical guidance on selection criteria and recommended combinations.

The Vercel Breach Started with a Roblox Cheat Search — Personal and Corporate Defenses Against Lumma Stealer

The Vercel breach began with a Context.ai employee searching for Roblox cheats and getting infected with Lumma Stealer. OAuth tokens for Google Workspace leaked, and an over-privileged third-party SaaS integration carried the attacker all the way into Vercel. This post lays out what individuals should be careful about and what organizations can actually do.

Opening a Folder Can Run Malware — A Safe Demo of Developer-Targeted Attacks Disguised as Job Recruiting via VS Code folderOpen and npm Lifecycle Scripts

Attacks that lure developers into running malware-laden repositories under the guise of recruiting or coding tests are on the rise. The steps you take right after git clone (opening the folder in an editor, running npm install) can themselves be the code-execution trigger. I built a demo repository to reproduce that safely on localhost, and here are the design decisions and defenses worth taking away.