Announced 15 Oct 2026 · Sources checked
What Ledger announced at Op3n on 15 October
Ledger published two primary write-ups on 15 October 2026: a product blog post from Ian Rogers, its Chief Human Agency Officer, and a longer Academy explainer on Agent Intent. Both say the feature launched the same day at Ledger Op3n in Paris, alongside reservations for Ledger Atom, a new device that ships in 2027 and is not required to try Agent Intent now.
The pitch is short. AI agents are starting to act like co-workers: they watch inboxes, prepare payments, rebalance positions and call APIs. Giving an agent that kind of reach usually means sharing credentials or keys. Ledger's answer is the line it repeats across the posts: agents propose, humans approve, hardware enforces.
Agent Intent is the piece that makes the waiting bearable. Instead of an agent freezing until you are at your desk with a signer plugged in, the agent keeps working and drops every approval request into one queue. You review when you are ready. Ledger says the product is free, live for everyone and in beta.
How Agent Intent works: propose, queue, approve on hardware
Ledger's developer guide describes the flow in three steps. You point an agent at the guide so it installs the official Ledger skill and creates its own identity key pair. You open the registration link, sign in with Ledger Sync, connect your signer and compare the Agent ID fingerprint on the device with the one the agent showed you. After that, proposals arrive as review links and wait in your queue.
What shows up in the queue is a proposal, not a finished transaction. The agent can prepare a swap, a payment, a staking rebalance or another supported action within the parameters you set. Clear Signing and Transaction Check are designed to show the amount, recipient and fees on the Secure Screen. You approve or reject with a physical confirmation on the device. If you reject, Ledger says the transaction is not signed or broadcast.
The agent signs each proposal with its own identity key so you can see which registered agent sent it. That signature identifies the agent; it does not approve the on-chain action. Ledger says a registered agent is designed not to register another agent, change the registry, sign a transaction or broadcast one without you.
What you need and which agents are supported
Hardware support is limited to up-to-date Ledger signers: Nano Gen5, Flex and Stax. The original Nano S is out. You need Ledger Sync (or to turn it on during onboarding), the Ledger Sync device app at version 1.3.0 or later according to the Academy post, and a browser that speaks WebHID or Web Bluetooth. Ledger names Chrome, Edge and Brave; Safari and Firefox do not work for the connection.
On the agent side, Ledger lists background agents that can keep working while you are away: OpenClaw, Hermes, Grok Bot, Muse and Dots. Coding agents such as Claude Code, Cursor and Codex are presented as a fit for the wider Agent Stack, including the Wallet CLI and Device Management Kit skills. Venice AI is mentioned for people who want private, model-agnostic routing so wallet details do not flow through a training-oriented provider.
That mix matters because agent wallets are no longer a crypto-only niche. Payment rails, API micropayments and shopping checkouts are already being wired to agents. American Express, for example, has published an Intent Intelligence Framework for agent purchases, and JPMorgan has said it will block outside AI agents from Chase accounts over liability concerns. Ledger is betting that hardware confirmation is the missing control for the people who still want agents to touch money.
Where it sits in Ledger's Agent Stack
Agent Intent is not a standalone product. Ledger places it on top of Agent Stack, an open toolkit it says has been live since July 2026, and inside a 2026 AI Security Roadmap published in April. Already-shipping pieces include Device Management Kit skills for connecting software to a signer, Transaction Check for scam flags, Clear Signing on the Secure Screen, agent skills and a CLI secured through the Ledger Keyring Protocol, FIDO2 security-key sessions and an OpenPGP app for decrypting secrets with physical confirmation.
Still ahead, according to the Academy post, are Agent Policies and Proof of You. Policies would let you pre-authorize rules such as a daily spend cap or a short allowlist of contracts, so in-bounds actions need no per-transaction tap. Proof of You is framed as a hardware-backed human check that is harder for a bot to fake. Neither is claimed as available in the Agent Intent beta itself.
Ledger Atom, revealed the same day, is the long-term home for the whole stack. Reservations for a Founders Edition at $849 are open, with an estimated Q2 2027 ship date that Ledger says is not a commitment. The important practical point for this story is the opposite one: Agent Intent is designed to run on the signer you may already own.
Demos: Grok Bot, Venice AI, OpenClaw and Pay.sh
Rogers' blog walks through four demos from the Op3n stage. In the Grok Bot example, the agent watches for an expected invoice, prepares a payment and parks it in Agent Intent until Rogers approves on a Nano Gen5. With Venice AI, an agent shops, fills a cart and only completes checkout after a device confirmation, with the privacy angle that personal details stay away from the model provider.
OpenClaw is the multi-agent case. Several sub-agents ask for different things at once: delete files, deploy code, move money. Agent Intent collapses those requests into one queue so approvals do not scatter across chat windows. Pay.sh, described as a Solana Foundation and Google Cloud payments path for agents paying APIs with stablecoins, is the autonomy ladder: software keys first, then Ledger for hardware keys (but the agent waits), then Agent Intent so the agent keeps working while payments wait in your queue, then a once-approved budget rule for in-bounds micropayments.
Those demos are vendor stagecraft. They show the intended product shape. They are not independent security audits, and Ledger itself labels Agent Intent as early experimental development provided without warranty.
Beta limits and what the agent still cannot do
Ledger is unusually blunt about the beta. Supported actions, behaviour and URLs can change without notice. Transactions you approve are real on-chain transactions and are generally irreversible. Until the beta ends, it tells users to connect Agent Intent only to non-critical workflows and low-value accounts.
Safety guidance in the Academy post is practical: trust the Secure Screen, not the agent's prose description; read the full recipient address, not just the first and last characters; never share a Secret Recovery Phrase with an agent; assume a compromised agent can still propose a transaction that looks reasonable. The control Ledger claims is your review, not the agent's good behaviour.
The docs also list operational constraints that will matter to people wiring this into production: you need an existing or newly activated Ledger Sync setup, the Sync device app installs during onboarding, and the browser must expose WebHID or Web Bluetooth. Remote autonomous agents on a VM are the recommended pattern, because Agent Intent is built for propose-anytime, approve-later rather than for a laptop that must stay open.
Why hardware approval matters for agent wallets
Ledger's Academy post spends time on why the risk is not theoretical. It cites a February 2026 report of an AI agent that tried to delete emails after being told to seek approval, and a security audit of a marketplace for agent add-ons that found hundreds of malicious skills, some disguised as crypto wallet tools. Prompt injection can turn a routine file or email read into a hidden instruction. None of that means agents are useless; it means final authority should sit somewhere the agent cannot rewrite.
That framing sits next to a wider industry fight over who is liable when an agent spends money. JPMorgan's refusal to let outside agents into Chase accounts is one answer. Shopify's Shop MCP checkout path for any agent is another, aimed at commerce rather than self-custody. Ledger is selling a third option: keep self-custody, let the agent do the legwork, and make the signer the place where a human still has to press a button.
Whether that holds up depends on implementation details that blogs cannot prove: how faithfully Clear Signing renders complex contract calls, how hard it is to social-engineer a rushed approval, and how Agent Policies will encode real spend limits without creating a new attack surface. Those answers will come from audits, incident reports and user experience after the beta, not from the launch posts.
What is still missing
Agent Intent is a product launch with clear primary documentation, but several important pieces are still future tense. Agent Policies and Proof of You are described as next layers, not shipping features of the beta. Ledger Atom is a reservation with a 2027 estimate. The partner list on stage is selective; a full catalogue of which on-chain actions are supported on day one is not in the posts we read.
There is also no independent timing or security evaluation in the announcement package. Ledger says the feature is experimental and without warranty. For readers, the honest next checks are the developer portal changelog, any published threat model for the intent queue, and whether Agent Policies arrive with hardware-enforced caps that match what the marketing describes.
Until then, the concrete claim you can verify from the primaries is narrower: a free beta exists, it runs on named Ledger devices, it queues agent proposals for human approval on a Secure Screen, and Ledger says private keys are designed to stay on the signer.
Common questions
Do I need Ledger Atom to use Agent Intent?
No. Ledger says Agent Intent works on compatible signers you may already own (Nano Gen5, Flex and Stax). Atom is a separate 2027 reservation announced the same day; Ledger calls it the best fit later, not a requirement for the beta.
Can the AI agent move funds on its own?
Ledger says a registered agent is designed not to sign or broadcast without your approval on the signer. Future Agent Policies may allow in-bounds actions after you approve a rule once; that layer is described as upcoming, not as part of the beta claim we verified in the launch posts.
Is Agent Intent free?
Ledger says it is free to use and live for everyone while in beta. Network fees and third-party provider fees can still apply to the transactions you approve.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. The cover places the companies’ own logos, taken from public-domain or openly licensed files credited in the caption, on a plain background. It is not a product photograph, a screenshot, or an endorsement.
Our editorial standards
