
OWASP Top 10 for LLM Apps 2026: Excessive agency risk on the rise
While prompt injection and data disclosure remain concerns, excessive agency climbed the list — not surprising with recent security incidents.
Build system hardening is the practice of securing the infrastructure, tools, and workflows involved in compiling, linking, and packaging software. It consists of implementing security controls that prevent unauthorized access, reduce the attack surface, and ensure the integrity of the software build process.
This process applies to CI/CD pipelines, build servers (e.g., Jenkins, GitLab, GitHub Actions), and associated systems that convert source code into deployable software.
Build systems are a prime target for attackers seeking to compromise software at its source. A successful attack can inject malicious code into trusted outputs, bypass security controls, and impact thousands of downstream users.
Hardening these systems:
Hardening involves implementing layered security controls across five key domains:
Practice | Focus Area | Key Difference |
|---|---|---|
Secure Build Environments | Physical and infrastructure-level isolation | Hardening includes policies, IAM, integrity, and monitoring |
CI/CD Pipeline Security | Workflow and process protection | Build system hardening focuses specifically on build components |
Runtime Security | Protects deployed software | Build hardening prevents threats before deployment |

While prompt injection and data disclosure remain concerns, excessive agency climbed the list — not surprising with recent security incidents.

The post-mortems of two compromises by rogue AI agents show that security teams need to focus on guardrails, not the AI model.

Here's how to use Spectra Analyze to hunt for malicious SVGs, from setting up queries and evaluations of samples to tips for investigation.