The challenge
What did the client need?
An enterprise software company had adopted AI-assisted development across several teams, and leadership could not say whether it was making the code less secure. It asked ioSENTRIX to measure the effect in three environments: a mature platform, a moderately mature system and an application built entirely with AI tools.
Engagement details
- Client
- Enterprise software company
- Environments
- A mature platform, a moderately mature system and a fully AI-native application
- Scope
- Pull requests, static analysis, release-stage scanning and CI/CD pipelines
- Alignment
- ISO/IEC 25010, NIST SSDF and OWASP SAMM
- Remediation
- DevSecOps roadmap delivered
- Delivered by
- ioSENTRIX application security team
Our approach
How did ioSENTRIX approach it?
The question was not whether one scan finds a bug, but how risk changed across environments over time. Answering it took analysis at the pull request, at release and in the pipelines, and comparing several tools against each other.
What we found
What did ioSENTRIX find?
AI sped up development, and in the weakest pipelines it sped up vulnerabilities too.
Theme · AI-nativeFinding 1
The AI-native app doubled its vulnerabilities
In the fully AI-native application, complexity and oversized functions surged, duplication rose and maintainability fell. Pull-request vulnerabilities doubled, with high-severity issues rising sharply.
Business impact. Without governance, AI-native development was building technical debt and security exposure at the speed of the AI.
Theme · ToolingFinding 2
No single tool saw everything
Pipelines differed in how they enforced scans, with missing dependency (SCA) coverage and limited container controls. Across the security tools, coverage and severity ratings diverged.
Business impact. Vulnerabilities could pass through one stage unseen, and leadership had no single view of the real risk.
Where AI-assisted development let risk through
Risk path · from code to release
- 1AI writes codevelocity rises
- 2Gates unevenscans not enforced
- 3Tools disagreegaps at release
Outcome: Risk scales (with development speed)
The impact
What changed as a result?
We found that mature teams could adopt AI safely behind strong pull-request gates, that uneven pipelines let AI expose hidden weaknesses, and that in the AI-native application, pull-request vulnerabilities doubled.
3
environments compared, from mature to AI-native
2x
pull-request vulnerabilities in the AI-native app
0
security tools that covered every environment on their own
What changed after the assessment?
What we found in each environment, and what we recommended| Environment | What we measured | What we recommended |
|---|
| Mature platform | Quality stable; security findings up modestly, mainly in authentication and authorisation logic | Adopt AI behind strong pull-request security gates |
|---|
| Moderately mature | Security findings up significantly in service and orchestration layers; tools disagreed | Pipeline fixes and a map of high-risk modules |
|---|
| AI-native app | Quality fell; pull-request vulnerabilities doubled; high-severity issues rose sharply | A governance roadmap with layered controls |
|---|
| CI/CD pipelines | Uneven scan enforcement, missing SCA coverage, limited container controls | Fixes mapped to pipeline stages, repository types and vulnerability classes |
|---|
✓ Verification: Roadmap delivered; adoption owned by the client.