The operating system for autonomous security research.
Verity is an autonomous AI red team, bug-bounty operations center, security-research laboratory, and evidence vault in one governed platform. Connect the applications, APIs, repositories, and programs you are authorized to test. Verity researches them continuously, proves real vulnerabilities through bounded, authorization-bound execution, and carries each verified finding from discovery to retest. An AI saying “this looks vulnerable” is never treated as a finding.
Machine-enforced authorizationBounded real executionIndependent evidence verificationHuman interlocksDurable custody and replay
a security team that never stops — inside strict controls
Autonomous security research, made governable and reproducible.
Connect what you are authorized to test, and Verity does the work of a continuously operating security team: it learns the rules, researches the target, attempts real exploitation within bounds, rejects anything it cannot prove, and keeps the evidence defensible from first request to final retest.
Knows exactly what it is allowed to test
Every run starts from verified ownership, normalized scope, permitted identities and techniques, request ceilings, and an expiring authorization grant. Missing, expired, revoked, or changed authority stops execution immediately — the stop button is part of the architecture.
Researches the target continuously
Multiple AI agents examine source code, application behavior, API contracts, browser flows, prior evidence, program rules, and attack surface, then form hypotheses about where real weaknesses are likely to be.
Tests hypotheses through bounded, real execution
Verity is not a scanner replaying signatures. It reasons about the target, tests through broker-gated transports, adapts its strategy, compares models, and uses reviewed skills — and it stops the moment authority expires or changes.
Separates real vulnerabilities from AI noise
A candidate becomes a finding only after target-reaching proof, an applicable safe-sibling negative control, independent critic review, replay, duplicate checks, and scope verification. Unprovable hypotheses are discarded, not presented.
Produces defensible evidence
Every meaningful action is bound to its authorization, identity, exact source, policy, scope, execution receipt, timestamps, SHA-256 digests, and custody references, so evidence can be replayed and reviewed by someone who was not there.
Manages the entire finding lifecycle
Discovery, evidence, independent verification, human approval, reporting, remediation, and authorization-bound retest live in one record — and, for operator-run bounty work, provider submission, dialogue, and outcomes too.
Built to operate across bug-bounty providers
Verity models the capabilities and limits of 17 provider ecosystems — directory, policy, scope, submission, readback, dialogue, retest, and outcome workflows. Each provider operation is enabled only once its exact authority is established; provider state comes only from authenticated provider evidence.
Stateful, two-actor authenticated testing
Use customer-provided actor and victim identities to prove object-level (BOLA, also called IDOR) and role-level authorization flaws, privilege escalation, and JWT token confusion that unauthenticated scanners cannot reach — without storing raw credentials.
Receives outside reports, too
Verity Inbound Reports gives external researchers, customers, partners, and internal teams an intake path, so reports you did not generate land in the same governed evidence and remediation workflow.
how a finding is made
A finding cannot become a report just because an AI says so.
Every stage, request budget, transport interlock, approval, and gate status is retained in the run record. Execution can recover without duplicating effects, and operators and customers retain immediate pause, cancel, and grant-revocation controls.
01Bind ownership, scope, identities, and the authorization grant.
02Research the target and form bounded hypotheses.
03Execute broker-gated proof with safe-sibling negative controls.
04Verify independently: replay, critic review, duplicate and scope checks.
05Require human approval before any external effect.
06Prepare the report with exact evidence and custody references.
07Track remediation and run the authorization-bound retest.
08Record closure with an immutable evidence receipt.
two principals, no inherited authority
Customer testing and third-party bounty work never share authority.
The owned_asset principal covers your assets, authorization, workers, evidence, and test identities, and runs only under a current explicit grant. The separate bounty_program principal covers operator-managed third-party programs, provider accounts, permissions, reports, and provider-originated outcomes. Neither side can borrow the other's credentials, authority, data, or evidence, and every unverified bounty cell remains disabled by default.
Security team, customer-owned assets
“Continuously test these five applications and APIs. Stay inside this scope. Use only these accounts. Do not perform destructive actions. Stop automatically if authority changes. Show me only independently verified findings, with exact evidence and retest results.”
Bounty operator, authorized programs
“Monitor these authorized programs, keep the rules and scope current, research permitted targets, prepare validated reports, require human approval before submission, and track every provider response and outcome.”
detection scope today
Six behavioral HTTP finders, with separate source-assisted analysis.
The behavioral HTTP registry covers BOLA (IDOR), reflected XSS, error-based SQL injection, template injection, path traversal, and JWT token confusion. Authorization bypass and SSRF are source-assisted classes and are not accepted by that HTTP registry. Browser, canary, identity, and isolation prerequisites remain explicit gates; a listed class is not a claim of release certification.
one governed system
Instead of buying five security products separately.
Verity is designed to replace — or coordinate — large parts of the stack security teams assemble today. It can also work alongside the bug-bounty platforms you already use, acting as the authorized research and operations layer for your programs rather than a replacement for their researcher communities.
Verity is not a vulnerability scanner; a wrapper around Burp Suite; a generic LLM security chatbot; a bug-bounty marketplace; a breach-and-attack simulation catalog; a pentest report generator; an agent allowed to attack anything it discovers.
Autonomous pentest platform
Agentic discovery and real exploit validation
Breach-and-attack simulation
Continuous, repeatable validation — of novel findings, not only a known-technique catalog
Bug-bounty program operations
Program rules, scope, reports, provider dialogue, and outcomes
PTaaS and retest services
Human-approved reporting, remediation tracking, and retest
Security-research operations stack
Models, reviewed skills, evidence custody, and replay
questions buyers ask
Frequently asked questions.
What is Verity?
Verity is an authorized autonomous offensive security and exposure validation platform: an AI red team, bug-bounty operations center, security-research laboratory, and evidence vault in one governed system. It researches the applications, APIs, repositories, and programs you are authorized to test, proves real vulnerabilities through bounded execution, and manages each verified finding from discovery to retest.
How is Verity different from a vulnerability scanner?
A scanner replays known signatures. Verity's agents reason about the target, form hypotheses, test them through broker-gated transports, and adapt. A candidate becomes a finding only after target-reaching proof, a safe-sibling negative control, independent critic review, replay, duplicate checks, and scope verification.
How does Verity know what it is allowed to test?
Every run is bound to verified asset ownership, normalized scope, permitted identities and techniques, request ceilings, and an expiring authorization grant. Missing, expired, revoked, or changed authority stops execution immediately.
How does Verity tell a real vulnerability from an AI hallucination?
A model saying a target looks vulnerable is never treated as a finding. Findings must survive executed proof, an applicable negative control, independent review, replay, duplicate checks, and scope verification, and every meaningful action is bound to its authorization, source, scope, execution receipt, and SHA-256 digests.
Which vulnerability classes does Verity test today?
Six behavioral HTTP finders cover BOLA (IDOR), reflected XSS, error-based SQL injection, template injection, path traversal, and JWT token confusion. Authorization bypass and SSRF are source-assisted classes, and the opt-in source-audit lane requires verified isolated infrastructure.
Does Verity submit bug-bounty reports on its own?
No. Bounty work runs in a separate operator-controlled principal, every provider operation stays disabled until its exact authority is established, and a human must approve before any submission. Provider responses and outcomes are recorded only from authenticated provider evidence.
How are customer testing and third-party bounty work kept apart?
Verity has two isolated principals. owned_asset covers a customer's assets, authorization, workers, evidence, and test identities; bounty_program covers operator-managed third-party programs and provider accounts. Neither can borrow the other's credentials, authority, data, or evidence.
Does Verity replace HackerOne, Bugcrowd, or Intigriti?
Verity is not a researcher marketplace. It is designed to act as the authorized research and operations layer for bounty programs, modeling 17 provider ecosystems, and can work alongside those platforms rather than replace their researcher communities.
reproducible measurement
Verification claims stay tied to their evidence and scope.
The public benchmark corpus, safe-corpus fixtures, judge prompts, and regression command are inspectable. Results are release evidence only for the exact source and configuration that produced them; they do not establish provider authority, a deployed bounty cell, or external certification.
Behavioral benchmark
Public corpus plus safe-corpus regression gate
Finding standard
Executed proof plus applicable negative control
Authority state
Missing, expired, or revoked grants fail closed
Provider operations
Each stays disabled until its exact authority is established
controlled enterprise pilot
Bring one application, two test identities, and one remediation cycle.