Hardware Security Protects Artificial Intelligence Agents

A new open-source toolkit combines hardware-enforced approval with artificial intelligence agents, ensuring sensitive actions require human confirmation while helping defend against prompt injection attacks.

Ledger has introduced an open-source toolkit that enables AI agents to interact with cryptocurrency wallets while requiring hardware-based human approval for sensitive operations. This approach makes it impossible for an AI agent to independently authorise any transaction.

This development addresses one of the key security issues facing agentic AI systems : prompt injection attacks. An AI agent can analyse wallet balances, create transactions and communicate with decentralized apps but all approvals are performed using the Secure Element chip installed in a hardware wallet. Private keys stay on the chip and are not exported. Every transaction needs to be reviewed and authorised manually using the display of the trusted hardware and a button.

This toolkit includes four open-source modules. Specifically, there is a device management toolkit for integrating hardware wallets into AI agents, a command-line interface (CLI) for personal wallets, an enterprise CLI, and a multisignature CLI. The device management toolkit is compatible with AI coding environments and command-line agent frameworks, providing integration between autonomous AI workflows and hardware wallet security.

The developers cited several examples of real-world incidents reported in 2026 where prompt injection, exposure of private keys, and parsing bugs led to the loss of cryptocurrency. It was explained by the developers that pure software-based protection measures could not prevent unauthorised transactions if an AI agent were compromised. Moving approval of transactions into hardware devices will create an additional security layer that hackers cannot bypass using software exploits.

Beyond cryptocurrency wallets, the hardware isolation can be applied for protecting digital credentials and authentication keys for use with online services. In future releases, developers plan to introduce features such as hardware enforcement of spending policies, programmable approval rules, and human verification proofs. The toolkit was released as an open-source project along with documentation and developer tools.