Security Review

AI-generated code can introduce security vulnerabilities. Some are bugs humans write too; some are unique to AI patterns.

AI-specific security concerns

Pattern-based vulnerabilities

AI learns from patterns—including vulnerable ones in training data.

Common issues:

• SQL/command injection
• Insecure deserialization
• Hardcoded credentials
• Missing auth checks
• Insecure defaults

Why: Models optimize for “code that looks right,” not “code that’s secure.”

Hallucinated security

AI may implement security incorrectly:

• Encryption using weak algorithms
• Authentication that doesn’t actually validate
• Authorization checks that can be bypassed
• Input validation missing edge cases

Code looks secure but isn’t.

Dependency risks

AI may suggest packages that:

• Don’t exist (hallucinated—could be typosquatted)
• Have known vulnerabilities
• Are unmaintained

Security review checklist

Input handling

• All user inputs validated
• Input length limits enforced
• Paths sanitized (no traversal)
• URLs validated

Authentication & authorization

• Auth properly implemented
• Tokens validated correctly
• Sessions managed securely
• Access controls on sensitive operations
• Authorization checked server-side

Data protection

• Sensitive data encrypted at rest
• TLS enforced
• Secrets not hardcoded
• No sensitive data in logs

Injection prevention

• Parameterized database queries
• Shell commands properly escaped
• No eval() with user input
• Template injection prevented

Dependencies

• All dependencies verified to exist
• No known vulnerable versions
• Lock files committed

Review by risk level

Code Type	Review Level
Any AI change	Skim for patterns, check inputs, verify deps exist
Security-sensitive	Full checklist, manual testing of boundaries
Auth/authz	Line-by-line, threat modeling, senior security review

Automated tools

Static analysis (SAST): Run on all code. Won’t catch all AI-specific issues, but catches common vulnerabilities.

Dependency scanning: Dependabot, Snyk. Check all deps exist before installing.

Secret scanning: Pre-commit hooks, repository scanning, CI/CD checks.

Team training

AI-specific:

• Common patterns AI gets wrong
• How to spot hallucinated security
• When to be extra suspicious

General security:

• OWASP Top 10
• Language-specific issues
• Secure coding guidelines

Incident response

When AI-generated code causes a security issue:

1. Treat like any security incident—don’t minimize because “AI did it”
2. Document AI involvement: Tool, prompt, what review happened
3. Root cause: Was this AI-specific? Would human have caught it?
4. Process improvement: What would have caught this earlier?

Post-incident: Update checklists, share learnings, consider tool-specific mitigations.

Resources

Videos

• Government Agents – Mark Myshatyn, Los Alamos - Security considerations for AI agents in government