Skip to content

[ML] Attack Path Correlation & Visualization for Security Findings #145

Description

@KolaSailaja

Summary

Implement an Attack Path Correlation & Visualization engine that connects related findings across scanners to identify potential attack chains and visually demonstrate how multiple vulnerabilities could be exploited together.

Motivation

PatchPilot currently displays findings as independent security issues.

However, attackers rarely exploit vulnerabilities in isolation. Real-world breaches often involve a sequence of weaknesses that combine into a complete attack path.

For example:

Hardcoded AWS Secret
        ↓
Cloud Access
        ↓
Vulnerable Dependency
        ↓
Remote Code Execution
        ↓
Sensitive Data Exposure

While each finding may be reported separately, security teams need visibility into how findings interact and contribute to larger attack scenarios.

Without attack path correlation, users may underestimate the impact of seemingly unrelated findings and struggle to prioritize remediation effectively.

Proposed solution

Introduce an Attack Path Engine that builds relationships between findings and generates visual attack chains.

Inputs

SAST findings
Dependency vulnerabilities
Secrets findings
Dependency graph
Affected files
Reachability data
Risk scores

Correlation Flow

Finding Collection
        ↓
Relationship Mapping
        ↓
Attack Chain Construction
        ↓
Attack Path Visualization

Example Output

{
  "attack_path_id": "AP-001",
  "risk_score": 95,
  "steps": [
    "Hardcoded AWS Secret",
    "Compromised Cloud Access",
    "Privilege Escalation",
    "Sensitive Data Exposure"
  ]
}

Backend

  • Build attack graph generation service.
  • Correlate findings across scanners.
  • Generate attack chains and path scores.
  • Persist attack path metadata.

Frontend

  • Interactive attack path graph.
  • Expand/collapse attack chains.
  • Highlight critical attack paths.
  • Visualize attack progression.

Evidence Pack Integration

Include:

attack-paths.json
attack-path-summary.txt
attack-graph-report.json

in generated evidence packs.

ML tier (if applicable)

  • Tier 1 — Triage (severity ranking, deduplication, false positive classification)
  • Tier 2 — Predictive (fix success prediction, exploit scoring, pattern clustering)
  • Tier 3 — Autonomous (LLM patch generation, self-healing pipeline)
  • Not ML-related

Alternatives considered

  1. Treat findings independently.

    • Rejected because real-world attacks often chain multiple vulnerabilities together.
  2. Prioritize only by severity.

    • Rejected because attack impact depends on relationships between findings, not just severity.
  3. Static attack templates.

    • Rejected because dynamic correlation can identify project-specific attack chains.

Acceptance criteria

  • PatchPilot correlates findings from multiple scanners into attack paths.
  • Users can visualize attack chains through an interactive graph.
  • Each attack path includes a calculated risk score.
  • Attack path artifacts are included in generated Evidence Packs.

Additional context

Modern security platforms increasingly focus on attack-path analysis rather than isolated vulnerability reporting. This feature would help PatchPilot move beyond traditional scanners and provide security teams with actionable insights into how vulnerabilities interact within a real attack scenario.

Metadata

Metadata

Assignees

Labels

mlML related issues

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions