Xcapit Labs / AiSec
Security Analysis Framework for AI Agents
Scan, detect, and remediate security vulnerabilities in AI agent systems. 35 specialized security agents, 250+ detectors, Docker-sandboxed execution, and compliance-ready reporting across 8 frameworks.
Capabilities
What AiSec Does
35 Specialized Security Agents
Each agent focuses on a specific security domain: prompt injection, data exfiltration, privilege escalation, supply chain, model poisoning, and more. Agents coordinate for comprehensive coverage.
Docker-Sandboxed Execution
Every scan runs in an isolated Docker container with resource limits, network restrictions, and filesystem isolation. Malicious payloads cannot escape the sandbox.
Auto-Remediation Engine
When vulnerabilities are detected, AiSec doesn't just report — it generates and applies fixes. Code patches, configuration changes, and policy updates are produced automatically.
Policy-as-Code Governance
Define security policies in code that are automatically enforced during scans. Custom rules, severity thresholds, and compliance requirements expressed as executable specifications.
AI-CVSS Scoring
Extended CVSS scoring system adapted for AI-specific vulnerabilities. Accounts for model access, data sensitivity, agent autonomy level, and cascade impact across agent systems.
SARIF & CI Integration
Export findings in SARIF format for IDE integration. GitHub Actions, GitLab CI, and Jenkins plugins enable security scanning as part of every pull request.
Validation
Proven in Production
OpenClaw Audit: 63 Findings
AiSec's comprehensive audit of the OpenClaw AI agent framework uncovered 63 security findings — 4.2x more than traditional scanners like Snyk or Semgrep alone.
31 Correlation Rules
Cross-agent correlation engine that identifies compound vulnerabilities. Individual findings that seem minor can combine into critical attack paths.
8 Compliance Frameworks
Built-in policy packs for OWASP AI Top 10, NIST AI RMF, EU AI Act, ISO 42001, ISO 27001, GDPR, SOC2, and MITRE ATLAS. One scan, all frameworks.
4-Minute Average Scan
Parallel agent execution with intelligent scheduling. Full security analysis of an AI system completes in under 4 minutes on average.
Our Journey
From Consulting to Framework
AiSec evolved from years of security consulting into a systematic framework for AI agent security analysis.
Security Consulting
Years of cybersecurity consulting for enterprises, building internal tools for vulnerability assessment, penetration testing, and compliance auditing.
Internal Tooling
As AI agents proliferated, existing security tools fell short. Built internal tooling to assess AI-specific vulnerabilities: prompt injection, data leakage, and agent autonomy risks.
Framework & OpenClaw
Formalized internal tools into the AiSec framework. The OpenClaw audit — finding 63 vulnerabilities where traditional scanners found 15 — validated the approach.
Open Source & SaaS
Released as open source with a SaaS platform for continuous monitoring. Enterprise customers run AiSec in CI/CD pipelines for automated security gates.
Security-First Architecture
AiSec is built for high-throughput parallel scanning with defense-in-depth isolation.
Orchestration engine for 35 security agents. Django REST API for scan management, finding aggregation, and compliance reporting.
Each scan runs in an isolated Docker container. Kubernetes orchestrates parallel execution with resource limits and network policies.
Falco with eBPF probes monitors container behavior in real-time. Detects anomalous system calls, network connections, and file access patterns during scans.
Roadmap
Vision 2026
AiSec is becoming the standard security scanner for AI agent systems, integrated into every CI/CD pipeline.
Use Cases
Who Uses AiSec
DevSecOps Teams
Integrate AiSec into CI/CD pipelines for automated security gates. Every PR that modifies AI agent code gets scanned before merge.
Compliance Teams
Generate compliance reports across 8 frameworks with a single scan. Map findings to specific requirements in OWASP AI Top 10, NIST AI RMF, and EU AI Act.
AI Agent Developers
Scan your agent systems during development. AiSec identifies vulnerabilities before they reach production and generates fix suggestions.
FAQ
Frequently Asked Questions
How is AiSec different from traditional security scanners?
Traditional scanners (Snyk, Semgrep) focus on known code vulnerabilities. AiSec's 35 specialized agents understand AI-specific attack vectors: prompt injection, data exfiltration through model outputs, privilege escalation via tool use, and supply chain attacks on model weights. The OpenClaw audit found 4.2x more issues than traditional tools.
What AI frameworks does AiSec support?
AiSec scans any AI agent system regardless of framework — LangChain, CrewAI, AutoGen, custom implementations. It analyzes the security boundaries, tool access patterns, and data flows rather than targeting specific framework APIs.
Can AiSec scan production systems?
Yes. AiSec supports both static analysis (source code scanning) and dynamic analysis (runtime behavior monitoring via Falco/eBPF). For production, it runs in observation mode without affecting agent performance.
How does auto-remediation work?
When a vulnerability is detected, AiSec generates a remediation plan including code patches, configuration changes, and policy updates. The remediation is proposed as a PR in your repository, with full explanation of the vulnerability and the fix.
Ready to secure your AI agents?
Whether you're building, deploying, or auditing AI agent systems — AiSec provides the comprehensive security analysis you need.