Smart-contract bug-bounty auditing is Hunt's v1 wedge because on-chain settlement closes the loop cleanly. The same machinery — sealed inference, multi-specialist competition, on-chain per-domain reputation — applies to the appendix domains below, but they support the main audit story rather than replace it.
The same v1 primitives fan out into four distinct adversarial-AI
surfaces. Smart contracts is the load-bearing v1 vertical with a settled
cryptographic proof on Aristotle mainnet. The appendix domains reuse
lib/credential.js's attestation primitive byte-for-byte; only the canonical class
strings differ.
Protocols seal Solidity code. AI hunter agents race through 0G
Sealed Inference to find vulnerabilities. v1 relays validated
findings through
teeSigner; the contract
ecrecovers the digest on-chain. Per-CWE reputation accrues per hunter, per
class.
Defends citizens against AI denials issued by health insurers in 1.2
seconds without individualized review. Same architecture as v1;
specialists analyze the denial letter for medical-necessity
misapplication, coding errors, prior-auth overreach,
network-adequacy violations, ERISA defects, and
state-external-review misclassification.
Live empirical validation 2026-05-13:
a real Sealed Inference call against the synthetic denial returned
strict JSON with real C.F.R. cites and TEE attestation validated by
broker.inference.processResponse.
d86b8797…f994
validated; 4/6 defect classes surfaced with real C.F.R. cites
For elderly, retired, and physically challenged citizens navigating SSDI, SSI, Medicare reconsideration, and VA claims. The largest adjudication backlog in any US administrative system, where attorneys are economically unavailable to most pro-se claimants. Specialists analyze denials for Listing misapplication, RFC errors, vocational-expert misclassification, and four more defect classes mapped to 20 C.F.R.
The cooperative face of the primitive. Gives patients better verifiable AI reads of their own records than they could otherwise afford. Specialists surface questions to ask your physician and second-opinion flags — never a diagnosis. Scope-locked to FDA Jan 2026 CDS enforcement-discretion guidance.
The vertical story now sits on reusable protocol pieces: developer primitives, a public AI receipt log, a cross-chain reputation view, and a partnership playbook for institutions that want their own Hunt instance.
Digest, class hashing, ECIES envelopes, attestation signing,
verification, and Notary helpers extracted into
packages/sdk/ with five runnable examples.
Users can notarize an AI transcript on Aristotle without putting the transcript on-chain. Contract stores content hash, model digest, domain, timestamp, and wallet provenance.
HuntReputationOracle wraps the live Hunt reputation
ledger into stable reads for apps, bridges, and future consumers
of per-domain model performance.
A deployment model for regulators, audit desks, consumer advocates, and ecosystem partners that need verifiable AI over a domain-specific class registry.
lib/cwe.js,
and the first non-Solidity bounty in each vertical has posted to
Aristotle mainnet —
bounty #23 (insurance, ERISA / medical-necessity / prior-auth),
bounty #24 (medical, pathology-borderline / oncology-staging — records-reader, NOT a diagnosis),
bounty #25 (benefits, SSDI medical-listing / RFC / treating-physician-opinion-weight).
Specialist hunters for these domains are not minted yet
(the verifier wallet's PK is held by the operator and the
appropriate human-in-the-loop partnerships — NOSSCR-credentialed
representatives for benefits, CLIA-certified pathologists for
medical — need to be in place before a non-operator hunter
races). The v2 bounties sit Open until expired-and-refunded by
expireBounty after the settle window; the
BountyPosted events are permanent on-chain proof
that Hunt's CWE registry now spans four domains, not just
smart-contracts. Bounty #3's strict verifier remains the
load-bearing cryptographic receipt for the v1 (smart-contract)
lifecycle: signer, race window, and supplied Sealed Inference
modelDigest must all match. Each vertical's
README.md
documents this explicitly — including an honest v1 privacy caveat
(the shared hunter-network-key gap that v2's per-hunter ECDH envelope
closes).