Skip to main content
Verity180Governed autonomous security research
ProductSecurityCompareControlled pilot
Sign inRequest pilot
ProductSecurityCompareControlled pilotSign in
Request controlled pilot

cryptographic isolation and safety

Every request is governed at the moment it leaves Verity.

Ownership, principal isolation, authorization, scope, method, rate, budget, transport health, and kill switches are checked immediately before execution. Ambiguous or stale state blocks closed.

Request security reviewOpen Hunter console
pre-egress interlock chainFail-closed
Gate 1Cryptographic Ownership & GrantDNS TXT, HTTP, or GitHub App proof plus an active SHA-256 policy snapshot
Gate 2ScopeOracle & NonDestructiveGuardRFC 1918, link-local, and metadata denial plus read-only proof verbs
Gate 3RequestInterlock & EgressBrokerAtomic request and cost budget, per-target RPS, and kill switch
Gate 4Optional gVisor Boundary & Redacted ManifestSource audit stays disabled without verified isolation; persisted evidence is scrubbed before storage

Cryptographic principal isolation and authorization

Customer-owned assets (principal_type = owned_asset) are strictly isolated from AuraOne-operated bounty work (principal_type = bounty_program). Every run references a policy snapshot and active authorization grant; expiry or revocation fails closed.

Reference-only secret handling

Hunter stores secret-manager references and encrypted session-vault references, never raw credentials, cookies, tokens, or browser state in product records or outbound reports.

Brokered execution safety

Enabled HTTP effects traverse hunter-egress-broker and RequestInterlock with DNS and IP rebinding blocks, redirect authorization, safe-method controls, atomic budgets, and kill switches. Browser effects remain disabled unless their separate authority and runtime prerequisites pass.

Opt-in gVisor source-audit boundary

The source_audit lane is disabled by default and is not production-certified. When separately enabled, it requires a verified Linux KVM host with the gVisor runtime, no target egress, a read-only root, dropped capabilities, and bounded resources; otherwise proof execution refuses to run.

Governed severity and cross-family critic

Severity comes from a deterministic classifier backed by cross-family adversarial verification via Kiro Prism. Unproven claims are discarded, and model output alone can never inflate severity.

Tamper-evident evidence manifests

Evidence is tenant-scoped, redacted before persistence, encrypted, retention-classed, and verified by SHA-256 on write, read, report generation, and export.

Strict tenant and disclosure boundaries

Customer identity is derived from authenticated context and enforced in route predicates, compound foreign keys, object keys, and audit records. Customer findings never enter any external disclosure channel.

Verity180Governed autonomous security research

The operating system for autonomous security research: authorization-bound testing, independently verified findings, defensible evidence, and retest.

Product
  • Product
  • Security
  • Compare
  • Controlled pilot
Unified console
  • Hunter overview
  • Inbound Reports
  • Sign in
© 2026 Verity Inc. All rights reserved.Authorized scope, verified evidence, human interlocks