Trust & safety

Security model

Who can move money, how secrets are kept, and what the AI is allowed to do.

Who can move money

Nobody, directly. Money moves only when an event (activation, an accepted delivery, a deadline) fires a rule in proved terms. The engine turns that into PayPal effects in a transactional outbox: one effect in flight per promise, each with a PayPal-Request-Id so a retry after a crash can never move money twice. A ledger mirror check freezes the promise if what PayPal confirmed ever diverges from what the rules decided.

Controls

ThreatControl
Account takeoverConvex Auth sessions, password policy (10+ characters, letters and digits), sign-in rate limiting
Reading other people's promisesEvery function checks participation; outsiders get NOT_FOUND, so ids leak nothing
Changing terms after approvalApprovals bind to termsHash (terms, both parties, fee, engine version); re-checked at funding
Redirecting payoutsPayout email locked at approval; only an arbiter can change it, journaled
Faking "verified"Only server code writes verification results; funding re-checks them
Double paymentsOutbox with idempotency keys, a crash reaper, one effect per stream at a time
Authorization expiryThe verifier enforces PayPal's window; an expiry guard captures to escrow a day early; just-in-time reauthorization
Stray paymentsMoney PayPal takes on a closed or replaced checkout is returned automatically
Funds that later failPayouts wait while a capture is PENDING at PayPal
Forged PayPal webhooksVerified with PayPal's API, de-duplicated by event id; webhooks never move money by themselves
Prompt injectionUntrusted wrapping, structured outputs, a manipulation flag, challenge windows, human arbiters
Rogue AI agentsDeterministic guard, caps, allowlists, velocity limits, expiry; AI can only escalate
Stolen tokensAPI keys, agent tokens and invites are 256-bit random, shown once, stored as SHA-256, scoped, expiring and revocable
Floods and abuseToken-bucket rate limits on compile, create, uploads, invites, verify, agent spends and the REST API
SSRF through webhooksEndpoints must be public HTTPS hostnames; private ranges, metadata hosts and trailing-dot tricks are blocked; redirects aren't followed
Tampering with historyHash chain plus Ed25519 signatures plus signed Merkle checkpoints
Clickjacking and look-alikesframe-ancestors 'none', a strict CSP, an official-host warning, same-site-only redirects

What the AI can and can't do

CanCan't
Draft promises and rules from plain EnglishChange a promise to make rules pass
Repair rules the verifier rejectedFund, release, refund or pay anything
Check a delivery against agreed criteriaAccept a delivery on its own (a pass opens a challenge window)
Screen an agent's purchase against its purposeApprove a purchase the deterministic guard blocks

Verify the history yourself

  1. Fetch the public keys: https://befitting-mosquito-713.convex.site/.well-known/kept-keys.json.
  2. Download a promise's proof pack (GET /v1/contracts/{id}/proof) or open the promise and press Verify in browser.
  3. Or audit offline: npx @kept/mcp verify-proof pack.json re-derives every hash, checks every signature and re-runs the verifier.
  4. Check that an entry is in the public log: GET /v1/transparency/proof/{entryHash}.

Known limits

  • Email ownership isn't verified at sign-up; invite links are single-use bearer secrets.
  • In the sandbox design, escrow-mode funds sit in the platform's PayPal balance; production should use PayPal's multiparty flows.
  • Arbiters are trusted people. Every arbiter action needs a reason and is visible to both sides.
Improve this page on GitHub