Ledger unveils hardware-backed Agent Stack to prevent rogue AI transactions

Ledger has launched Ledger Agent Stack, an open-source toolkit that lets AI agents interact with crypto wallets and blockchain tools while keeping humans firmly in control of sensitive actions. 

Agents will be able to read balances, suggest transactions, prepare swaps, and draft operations, but will require explicit human sign-off on their physical hardware devices before execution.

The move comes amid a wave of AI integrations in some of crypto's most-used platforms. For instance, MetaMask is rolling out agentic capabilities that will follow a similar safety-first approach by requiring human approval to execute transactions. 

Ledger notes that it is preparing for a world where AI agents "become more autonomous." It is building the capability for its customers to deploy bots, but is taking a relatively conservative approach. 

While some wallets are already allowing bots to autonomously execute actions within user-defined rules, Ledger will only let agents propose actions, a move meant to protect users from AI mistakes as well as other potential risks.

"As the attack surface involving AI agents grows, losing defense strategies rely too heavily on software firewalls, LLM guardrails, and input filtering," a Ledger representative said. "Even big corporates like Starbucks and Meta are rolling back their AI agents due to privacy fears and rogue agents."

Ledger also noted that the AI "attack surface is growing faster than the defense," adding that 26.1% of all agent skills contain at least one security vulnerability. Likewise, AI proposals may also cut down on human error, which is attributable to about 60% of all security breaches. A hybrid AI-human approach can help mitigate risks from both compromised agents and human mistakes.

"By anchoring every signing decision in hardware, Ledger Agent Stack is designed to ensure that even if an agent's software environment is compromised, funds cannot move and sensitive information cannot be accessed without explicit human confirmation on your Ledger signer," it added.

Notably, Ledger is also pitching its security devices for users outside the crypto industry. The company noted its support for the open encryption standard OpenPGP means any developer can use their physical Ledger to encrypt API keys and AI agent credentials. Users can also use their devices like YubiKey security systems, requiring the device to log into accounts on GitHub, npm, 1Password, Discord and elsewhere. 

The Ledger Agent Stack includes a set of open-source Markdown instructions that teach AI bots how to wire Ledger hardware directly into their "agentic workflow." The release notes MoonPay and Shisa.ai have already added this feature to their products. 

Thursday's release is part of Ledger's 2026 AI Roadmap, which plans to introduce additional tools for agent identity and execution as well as "proof of human" attestation in the coming months. 

Founded in 2014, the Paris-based company is said to be exploring an initial public offering that could value it at more than $4 billion.