Never give an AI agent a wallet without a kill-switch.

Agent identity verification confirms that an AI agent trying to act or pay is the agent it claims to be, that it carries real authority delegated by a named principal, and that the action falls within that authority, all checked before the transaction executes. FLINT performs this verification and emits a signed record of the result.

FLINT is trust infrastructure for the agents that move money. It verifies an agent's identity and financial authority at the moment it acts, returns a verdict, and emits a signed verification record you can keep as evidence.

Six layers: the principal behind the agent, the agent's own identity, the provenance of its wallet, the scope of its authorization, its runtime environment, and its reputation across merchants. Together they answer one question: should this agent be allowed to take this action right now.

An agent can be entirely real and still act outside its authority. A compromised, misconfigured, or over-delegated agent is the new fraud surface. FLINT verifies the delegation chain that flows from a principal down to the action, so a genuine agent attempting an ungranted action is caught. Most controls confirm the agent is real; FLINT confirms the action is authorized.

Each record captures the agent, the principal, the authorized scope, the layer results, the trust score, and the verdict, signed with a hybrid scheme (ES256 plus ML-DSA-65) so it stays verifiable as quantum-capable adversaries emerge. Because it is signed, the record is tamper-evident evidence: you, an auditor, or a counterparty can later prove what was verified and what was decided.

Yes, and detecting that is the point. Because every verification is a signed record, a corrupted or drifting agent shows up as a broken signature or a changed behavioral profile on its next action. A security team can pull the records for every agent on its network and see at a glance which ones no longer verify.

Device fingerprinting identifies the browser, device, and session a request comes from, and it assumes a human is behind the screen. FLINT identifies the agent itself and the financial authority it carries: who delegated it, what it can spend, and whether this action is in scope. Device signal is one of our six layers, not the whole picture.

On-chain analytics trace funds after they move. That is forensic and backward-looking. FLINT verifies the agent's authority before money moves and produces a signed record at decision time. We sit in front of the transaction; on-chain analytics sit behind it. The two are complementary.

No. FLINT is rail-agnostic and holds no funds. It verifies the agent and returns a verdict; the rails move the money. We work alongside x402, Circle, Stripe, and the card networks rather than competing with them.

No. Blunt blocking is the problem we solve. Most businesses already lose more revenue to false declines than to fraud. FLINT returns a four-state verdict (allow, step-up, review, block) so good agents pass, ambiguous ones get a check, and only the unauthorized are stopped. The goal is to recover the revenue crude rules reject.

AI agent takeover is when an attacker gains control of an AI agent that can spend money, by hijacking the server or framework it runs on, poisoning its instructions, or stealing its credentials. The agent keeps its valid identity, so it passes checks that only ask who it is, but it now acts for the attacker. It is account takeover moved up to the agent layer.

A kill-switch lets the owner instantly stop an AI agent from transacting. With FLINT Sentinel, the verified owner can freeze the agent's passport from an out-of-band alert on Telegram or email, and a frozen agent is blocked from transacting anywhere FLINT is checked. Freeze is reversible.

Give the agent a verifiable identity (a FLINT agent passport), set explicit spending limits, watch for tampering with those limits, and keep an out-of-band kill-switch. FLINT Sentinel does the watching and the kill-switch: it alerts you when an agent's permissions are escalated and lets you freeze it with one tap.

No. FLINT does not patch the framework or prevent the breach itself. It catches the takeover the moment the agent tries to escalate its authority or move money, alerts you out-of-band, and lets you freeze it, so a compromise upstream does not become an unbounded loss. It is one layer of defense in depth.

FLINT Sentinel turns an agent passport into a portable anti-fraud tool. It watches an AI agent's permissions, logs every change, alerts the owner out-of-band when those permissions are escalated, and provides a one-tap reversible freeze.