ATLAST Protocol optionally anchors AI agent evidence on Base blockchain via EAS (Ethereum Attestation Service) — creating permanent, public, independently verifiable proof of agent behavior.
ATLAST's Evidence Chain Protocol (ECP) already creates tamper-proof SHA-256 hash chains. Blockchain anchoring adds three additional guarantees:
Important: Agent data (prompts, outputs, reasoning) NEVER goes on-chain. Only the SHA-256 hash is anchored. Your content stays private on your device. The blockchain only stores proof that the evidence exists and hasn't been tampered with.
Every action is recorded locally as a signed, hash-linked evidence chain.
At configurable intervals, the latest chain hash is submitted to EAS on Base.
An EAS attestation is created containing: agent DID, chain hash, timestamp, and record count.
Anyone can verify: look up the attestation on Base → compare hash against the local evidence chain → cryptographic proof of integrity.
| Feature | Off-Chain Only (ECP) | On-Chain Anchored |
|---|---|---|
| Tamper-proof | ✅ SHA-256 chain | ✅ SHA-256 + blockchain |
| Self-verifiable | ✅ | ✅ |
| Public proof | ❌ Private | ✅ Anyone can verify |
| Permanent | Depends on storage | ✅ Immutable |
| Cost | Free | ~$0.001/anchor |
| EU AI Act | ✅ Compliant | ✅ Gold standard |
Not every agent needs blockchain. Here's when on-chain anchoring adds value:
| Scenario | On-Chain Needed? | Why |
|---|---|---|
| Internal dev agents | Optional | Internal trust is sufficient |
| Financial transactions | Recommended | Regulatory audit trail |
| Legal/compliance | Recommended | Tamper-proof public timestamp |
| Agent marketplace | Recommended | Public verifiable reputation |
| Healthcare/medical | Recommended | Patient safety records |
Anchoring every individual ECP record on-chain would be expensive and unnecessary. Instead, ATLAST uses Merkle tree batching — an efficient cryptographic technique that allows thousands of records to be verified with a single on-chain transaction.
Here is how it works: multiple ECP records are collected into a batch. Each record's hash becomes a leaf node in a Merkle tree. The tree is built bottom-up by hashing pairs of nodes together until a single root hash is produced. Only this root hash is submitted to the blockchain via an EAS attestation. To verify any individual record, you only need the record itself plus a small number of sibling hashes (the "Merkle proof") — not the entire batch. This is logarithmically efficient: verifying one record in a batch of 1,000 requires only about 10 hash comparisons.
The result is that a single on-chain transaction costing fractions of a cent can anchor hundreds or thousands of evidence records, making blockchain-level verifiability practical even for high-volume agents. The Merkle proof for any individual record can be generated on demand and verified by anyone with access to the on-chain attestation.
Not all blockchain-based approaches to AI trust are equal. Some projects put AI model weights on-chain, others create AI agent NFTs, and some track inference results on decentralized networks. ATLAST takes a fundamentally different approach: only evidence hashes go on-chain. This design choice reflects several important principles.
Privacy preservation: Agent data (prompts, outputs, reasoning) stays private on the agent's local device. Only cryptographic hashes are published, revealing nothing about the actual content. This satisfies both privacy regulations (GDPR, HIPAA) and enterprise security requirements.
Scalability: Hash anchoring is orders of magnitude cheaper than storing data on-chain. An EAS attestation on Base costs a fraction of a cent, regardless of how much data the hash represents. This makes on-chain trust practical for production AI agents that process thousands of actions per day.
Chain agnosticism: While ATLAST currently uses Base (Ethereum L2), the hash-based approach is inherently chain-agnostic. Evidence hashes could be anchored on any blockchain — Solana, Bitcoin (via timestamps), or future chains — without changing the ECP protocol. This protects against chain-specific risks and ensures long-term interoperability.
As AI agent marketplaces emerge — platforms where users can browse, compare, and hire agents for specific tasks — on-chain evidence becomes the foundation of agent reputation. When an agent's evidence chain is partially or fully anchored on-chain, any marketplace participant can independently verify the agent's track record without trusting the marketplace operator. This creates a decentralized reputation system where trust is based on cryptographic proof rather than platform-specific ratings that can be gamed or manipulated. For marketplace operators, on-chain evidence reduces fraud risk and increases user confidence. For agent developers, it means their agent's hard-earned reputation is portable across marketplaces — not locked into any single platform.
Blockchain provides permanent, public, independently verifiable timestamps. ATLAST anchors Merkle roots on Base via EAS, creating an immutable public record.
No. Blockchain anchoring is completely optional. Evidence chains are fully valid without it. On-chain adds extra verifiability for high-stakes scenarios.
ATLAST batches evidence records into Merkle trees and anchors only the root hash. A single on-chain transaction covers thousands of evidence records, keeping costs to fractions of a cent per record on Base L2.
Yes. All ATLAST attestations are created via EAS (Ethereum Attestation Service) on Base blockchain. You can look up any attestation using the EAS Explorer, verify the agent DID, chain hash, timestamp, and record count directly on-chain. No special tools required — just a web browser and the attestation UID. The ATLAST CLI also provides a atlast verify command that automates this process.
Optional blockchain anchoring for maximum verifiability. Or start with off-chain — it's still tamper-proof.
Get Started with ATLAST →