ZeroWatch is built around deterministic evidence, careful scope boundaries, tenant isolation, and reports that separate observed facts from AI-generated analysis.
Scope boundaries
Passive and minimally invasive checks come first.
Verified ownership is required for higher-trust or expanded scanning modes.
Every report states what was checked and what was not checked.
Evidence pipeline
Scanner modules produce deterministic evidence before any AI processing.
Normalized findings are versioned and attached to the scan snapshot.
AI adds prioritization and remediation language but does not fabricate evidence.
Abuse controls
Per-target request budgets, concurrency caps, and cooldown windows.
Workspace-scoped audit logging for scans, active validations, report access, and AI usage.
Strict rate limiting, approval gates, and scheduled execution boundaries.
Data handling
Workspace data is tenant-scoped and protected with application context and database row-level security.
Agent and MCP responses are intentionally minimized and do not expose infrastructure or database internals.
Sensitive runtime secrets stay in managed secret storage and are not exposed through reports, prompts, or agent tools.
Framework alignment
OWASP ASVSOWASP Top 10OWASP API Security Top 10OWASP Top 10 for LLM ApplicationsNIST CSF 2.0NIST SSDFNIST AI RMF 1.0
What ZeroWatch does
Performs evidence-first external posture scanning against public web targets.
Supports verified-target active validation and live DAST on higher-trust plans.
Provides historical reports, diffs, exports, alerts, and audited agent workflows.
What ZeroWatch does not do
It is not a generic exploit launcher or unrestricted autonomous hacking tool.
It does not treat AI output as evidence when scanner evidence is missing.
It does not replace a manual penetration test for high-trust environments.