Ledger, the company best known for its crypto hardware wallets, has introduced a new product aimed at the growing intersection of artificial intelligence and digital assets. The Agent Stack, announced Tuesday, is a hardware-backed system designed to secure AI-initiated transactions while keeping a human in the loop.
What the Agent Stack does
The stack is built around a dedicated hardware module that signs transactions only after verifying they pass a set of programmable rules. Those rules can include spending limits, approval lists, and multi-signature requirements. The idea is to let AI agents — autonomous programs that trade, pay bills, or manage portfolios — execute routine tasks without giving them unfettered access to private keys.
Transactions are logged on a tamper-resistant ledger, and the hardware ensures that even if the AI is compromised, the keys remain inside the device. The company says the system is meant to balance the efficiency gains of automation with the need for human oversight, a trade-off that has become central as more financial services move toward autonomous agents.
Why hardware matters
Software-only security has a well-known weak point: if the computer running the AI is infected, the attacker can steal keys or manipulate the agent's behavior. Ledger's approach keeps the signing key inside a physical chip that communicates with the AI only through a narrow, verified channel. That means the AI can request a payment, but it cannot order the chip to sign something that violates the pre-set rules.
Ledger has been making similar hardware for cryptocurrency wallets for years, and the Agent Stack borrows the same secure-element technology. The difference is that the new system is designed to be integrated into AI orchestration platforms, not just crypto apps. The company describes it as a “trust anchor” between autonomous agents and the real-world assets they control.
Balancing automation and human control
The tension between letting AI act fast and keeping it on a leash is not new. But as AI agents start handling payments, subscriptions, and even contract negotiations, the stakes get higher. A bug or a jailbreak could drain an account in seconds. Ledger's solution lets human operators set guardrails in advance — for example, a maximum daily transfer amount or a whitelist of approved counterparties — and then step in only when the agent hits a boundary.
That design is a deliberate middle ground. Fully manual approval kills the speed benefit of AI. Full autonomy risks catastrophic errors. The Agent Stack tries to offer both, though it requires users to trust that the hardware's rule engine is correctly configured.
Ledger has not said when the stack will be available for general use or how much it will cost. The company is positioning it as a platform for developers, meaning the first implementations will likely appear in custom trading bots, automated payroll systems, or smart-contract interfaces. Whether the broader market will embrace a hardware add-on for AI agents remains an open question, but the underlying problem — how to let an AI spend money safely — is not going away.
