Ledger releases open-source stack for human-approved AI crypto transactions

Ledger SAS has launched Ledger Agent Stack, an open-source developer toolkit for AI agents that can inspect wallet data and assemble crypto transactions while leaving final approval to a person using Ledger hardware. The Paris company is positioning the release as security infrastructure for agentic AI workflows where autonomous software touches financial accounts, a use case that becomes risky when private keys or signing authority sit in software.

The product does not give agents the ability to move funds by themselves, according to Ledger. Agents can check balances, review transaction history and prepare actions such as sends, swaps and staking, but execution requires confirmation on a Ledger signer. The company said that design is meant to limit damage if an agent runtime, integration or surrounding software is compromised.

Ledger did not announce pricing, revenue impact or customer adoption metrics for Agent Stack. It named MoonPay Inc. and Shisa Inc. as partners that have already used its Device Management Kit to add Ledger support to their products.

What Ledger is shipping

Agent Stack includes several open-source components aimed at developers building AI workflows around wallets and treasury systems. Device Management Kit Skills provide instructions for large language model runtimes to connect with Ledger devices without requiring teams to rebuild wallet logic from scratch. The Ledger Wallet CLI gives programmatic access for balance checks, receiving assets and preparing transactions.

Ledger is also extending the model to institutional users. Its Ledger Enterprise CLI and Ledger Enterprise Multisig CLI apply similar controls to treasury operations, with policy enforcement tied to hardware security modules, according to the company.

The release also includes two device apps that move beyond standard wallet operations. An OpenPGP app is designed to encrypt agent secrets so they cannot be accessed without a Ledger signer connected. A Security Key app gates access to services including GitHub, npm, 1Password and Discord through the device, reducing the value of a stolen password on its own.

Ian Rogers, Ledger’s chief human agency officer, framed the launch as an extension of the hardware wallet model into AI-operated software. “Crypto wallets have protected billions on this standard for years,” Rogers said. “Ledger Agent Stack allows your agent to use these wallets just as easily as humans.”

The AI security angle

Ledger is grounding the launch in a straightforward claim: agents should be able to propose actions, but signing should remain hardware-bound and human-approved. The company cited research that attributes about 60% of breaches to human error, as well as a separate study that found roughly 26% of agent skills contain at least one vulnerability.

Those figures support Ledger’s argument, but they do not prove that hardware signing solves the broader security issues around AI agents. Agent Stack reduces exposure around key custody and transaction approval. It does not eliminate risks in prompts, third-party tools, off-chain permissions or flawed policies. For crypto teams testing agents in production workflows, the practical value is narrower and more concrete: keep the agent away from the private key and require a physical approval step for sensitive actions.

Agent Stack is the first release in Ledger’s 2026 AI roadmap. Ledger said further products covering agent identity, policy controls and a “proof of human” check are planned for later this year. Documentation is available at developers.ledger.com/agent-kit.

Founded in 2014, Ledger says it has sold more than 8 million hardware signers in over 165 countries. That installed base gives the company a distribution advantage if developers decide hardware approval should be part of agent workflows, though Ledger has not disclosed how many customers are expected to use Agent Stack.

This story draws on original reporting from SiliconANGLE.