Validating AI Agent Records with Akmon

Practical Summary and Purpose

Akmon is a tamper-evident evidence and verification layer designed for AI agents. It addresses the challenge of proving agent actions to regulators, security reviewers, or incident teams long after an event has occurred. By treating evidence as a portable, cryptographically signed record, Akmon allows third parties to verify the integrity and authorship of an agent's session offline, without requiring the original agent's environment or cloud services. This capability is particularly relevant for meeting high-risk logging obligations under the EU AI Act, which apply as of 2 August 2026. The tool focuses on the verification chain, ensuring that evidence remains immutable and attributable.

The core value proposition of Akmon is its ability to provide a self-contained evidence artifact. Unlike standard telemetry that lives in mutable process memory or unsigned spans, Akmon seals the session data. This allows for independent verification by auditors who may not have access to the original agent's infrastructure. By providing a clear, verifiable audit trail, organizations can demonstrate accountability for AI-driven decisions, which is a critical requirement for compliance in high-risk AI deployments. The tool is designed to be agnostic, sitting on top of any agent that emits OpenTelemetry, making it a versatile addition to existing AI governance stacks.

Prerequisites and Installation

To use Akmon, you need a system running macOS or Linux. The tool is built in Rust (1.88+) and can be installed via Homebrew or by downloading prebuilt binaries from the official repository. When downloading binaries, it is essential to verify the integrity of the files using the provided SHA256SUMS file. For the verification step, specifically when using openssl, you must ensure you are using OpenSSL 3.x, as older versions like LibreSSL (often found on macOS) do not support Ed25519 signatures. You can install the required version using brew install openssl@3.

The installation process is straightforward for both the primary CLI tool and the standalone agef-verify binary. The latter is specifically intended for auditors who need to verify bundles without installing the full Akmon suite. By separating the verification logic into a lightweight binary, Akmon ensures that the barrier to entry for third-party auditors remains low. Users should always prioritize verifying the checksums of downloaded binaries against the published SHA256SUMS to maintain the integrity of the toolchain from the start.

The Documented Workflow

The Akmon workflow transforms agent telemetry into a sealed, verifiable bundle. The process begins by importing an OpenTelemetry trace or using Akmon's own agent to generate a session. Once a session is captured, you generate a signing key, sign the bundle, and optionally attest an operator. The following sequence demonstrates the transition from a trace to an offline-verifiable proof:

# 1. Make a signing key. Keep the private half secret and hand out only the .pub.
akmon bundle keygen --out signer.pk8 --public-out signer.pub

# 2. Turn a real trace into a session, then a portable bundle.
akmon otel import trace.json --journal ./journal
akmon bundle export <session-id> --journal ./journal --output audit.akmon

# 3. Sign it.
akmon bundle sign audit.akmon --key signer.pk8

# 4. Verify integrity, the signature, and the capture level.
akmon bundle verify audit.akmon --verify-key signer.pub --require-signature

# 5. Prove it with openssl alone, no Akmon involved.
akmon bundle prove-openssl audit.akmon --verify-key signer.pub --out-dir proof
openssl pkeyutl -verify -pubin -inkey proof/pubkey.pem -rawin -in proof/statement.bin -sigfile proof/signature.bin

This workflow ensures that every step of the evidence lifecycle is accounted for. By generating a signing key, you establish a root of trust for the session. The signing process then binds this identity to the session data, creating a tamper-evident record. Finally, the ability to prove the signature using standard openssl commands provides a universal verification method that does not rely on proprietary software, fulfilling the requirement for cloud-independent and vendor-neutral evidence.

Understanding Capture Levels

Akmon distinguishes between two levels of evidence completeness, which are signed into the record to prevent misrepresentation. A full capture level is achieved when using Akmon's own agent, which records prompts, responses, and tool calls, enabling deterministic replay via akmon replay. Conversely, a structural capture level results from importing OpenTelemetry traces from other agents. While this provides the session's shape, metadata, and tool calls, it does not support byte-for-byte replay. Akmon enforces these distinctions; for instance, akmon bundle verify --require-capture full will fail if the record is only structural, ensuring that users cannot pass off imported traces as full recordings.

This distinction is vital for maintaining the integrity of the audit process. By explicitly labeling the capture level, Akmon prevents the overstatement of evidence quality. A full capture provides the highest level of assurance, as it allows for the deterministic reconstruction of the agent's actions. A structural capture, while less comprehensive, still provides valuable context for auditing. By signing these levels into the record, Akmon ensures that any verifier can immediately understand the limitations and capabilities of the evidence they are reviewing, preventing potential misunderstandings during compliance audits.

Limitations and Compliance

While Akmon provides a robust framework for evidence, it is not a compliance certification and does not guarantee adherence to any specific legal framework. It is intended to assist in producing evidence for standards such as the EU AI Act, the NIST AI Risk Management Framework, and SOC 2. Users must validate any mapping of AGEF evidence to specific controls with their own legal and compliance teams. Furthermore, Akmon is not a replacement for enterprise governance platforms like Microsoft's Purview or Azure Confidential Ledger; rather, it is designed to complement them by providing a portable, agent-aware, and cloud-independent verification layer that works across different environments.

The tool is specifically designed to be an evidence layer, not a comprehensive governance suite. It does not attempt to replace existing enterprise tools but rather to fill the gap where those tools may lack portability or agent-specific replay capabilities. By focusing on the core task of creating and verifying tamper-evident records, Akmon remains a specialized and effective tool for its intended purpose. Organizations should view Akmon as a component of their broader AI governance strategy, integrating it with other tools to ensure a holistic approach to compliance and security.