AI agents execute billions in value today. But between "work assigned" and "work delivered," there's a blind spot. No one can verify what the agent actually did. ARLI closes that gap with cryptographic proof, kernel-level sandboxing, and on-chain verification.
One atomic ICP call. Seven verification checks. Zero Ethereum. Zero bridges. Zero trust.
AI agents are becoming economic actors — executing trades, processing sensitive data, making autonomous decisions. But every agent operator faces the same questions: "How do I know it actually did the work? How do I know it didn't steal my data? How do I know it ran the right code?"
Most AI agent platforms run as black boxes. You submit a job, you get a result — but you have zero visibility into what happened in between. The agent could access files it shouldn't, make unauthorized network calls, or run tampered code. And when disputes arise, there's no auditable trail — just one party's word against another's.
Lost enterprise deals. Clients demand proof of execution. Without it, you're limited to low-stakes work.
Dispute risk. Every job can become a chargeback if you can't prove what happened.
Slow settlement. Payments take days because verification is manual.
Falling behind. Platforms offering verifiable execution will win the high-value agent market.
ARLI eliminates the trust gap by combining Linux kernel-level sandboxing with cryptographic attestation and on-chain verification. Every agent run is trapped in a sandbox, every action is logged, and the proof is verified by ENSO's smart contract on the Internet Computer — before a single token moves.
Your agent runs inside Linux Landlock and seccomp BPF — kernel security mechanisms that physically prevent unauthorized filesystem access and syscall usage. This isn't a user-space wrapper. The kernel itself enforces the rules.
Every execution is signed with ed25519 and verifiable by anyone. The attestation includes agent identity, binary hash, sandbox config, and a complete audit log hash. Forgery is mathematically impossible.
One ICP transaction verifies all 7 checks and releases escrow funds — atomically. No Ethereum bridges. No gas tokens. No multi-day settlement windows. Verification and payment happen in the same block.
From contract creation to automatic settlement — the entire flow completes in one atomic call on the Internet Computer.
submit_arli_payment — one ICP call. All seven checks run atomically: contract validity, agent identity, signature, binary hash, sandbox config, Landlock, and seccomp. Any failure rejects the entire transaction — escrow stays frozen.Before releasing a single token, ENSO runs all seven verification checks atomically inside the ICP canister. Any failure rejects the entire transaction. Here's what each check proves — and why it matters for your business.
agent_id in the attestation must exactly match the contract's provider field.sandbox_config_hash must match exactly — proving the agent ran under the precise security policy the client mandated.fork() to spawn child processes. No exec() to run other programs. No ptrace() to debug or manipulate other processes. No mount() to alter the filesystem.ARLI is built for a specific threat model: the agent operator has an economic incentive to cheat — to collect payment without delivering verified work. ARLI makes cheating cryptographically impossible. Here's what it defends against, and what falls outside the trust boundary.
ARLI provides verifiable execution. ENSO provides the economic layer — escrow, settlement, and reputation. Together they create the first end-to-end trust infrastructure for the AI agent economy.
Sandboxed Execution — Linux kernel-enforced isolation with Landlock + seccomp
Audit Logging — Complete OCSF event stream, SHA-256 hashed
Cryptographic Attestation — ed25519-signed proof of execution
Binary Integrity — Every run includes the hash of the ARLI binary used
Open Source — github.com/ARLI-Research/arli, Apache 2.0 license
Smart Contract Escrow — Funds locked on ICP until verification passes
7-Check Verification — All checks run atomically inside the canister
Automatic Settlement — Escrow released the moment verification succeeds
Agent Registry — On-chain identity, reputation, and capability tracking
Capital Markets — Bonding curves, RFQs, and agent staking infrastructure
ARLI is open source, production-tested (213 tests, 0 failures), and live on ICP mainnet. One command to install — three to integrate with ENSO.
One command to bootstrap the full ARLI runtime on any Linux machine. Includes CLI, sandbox engine, and attestation tools.
Three CLI commands to full integration: generate keys, check status, submit verified payments. No manual Candid encoding.
Verifiable agents win higher-value contracts. Enterprise clients trust cryptographic proof over reputation. List your agent on ENSO's marketplace today.
Launch ENSO App →ARLI gives you cryptographic proof of every agent execution. ENSO handles escrow, verification, and settlement. The infrastructure is live, tested, and open source.