Ledger Executive Ian Rogers: AI Can Research, Reason, and Suggest, But Actions Must Come from Human-Set Parameters
Ian Rogers, Chief Human Officer at Ledger, stated that AI can research, reason, and suggest, but actions must come from strategies or explicit approvals set by humans, which is the only feasible path for alignment.
He added during a Bloomberg Businessweek daily program that security needs to be redesigned for autonomous, probabilistic "beings" that assist in work. The company established this position in April, appointing a former Chief Experience Officer, with a roadmap requiring significant decisions to remain on hardware trust roots. Agents can check balances, analyze portfolios, and prepare transactions, but sensitive actions must be approved on a trusted screen of Ledger devices. Agents within the strategy can run automatically all day, while those outside the strategy require human confirmation. Governance is placed in an independent strategy engine outside the model, and the model cannot alter authorization rules.
On the product side, there is already a device management suite and agent stack, with integrations like MoonPay for "agent proposals, human approvals." Plans include hardware-bound agent identities and connecting human proof to secure chips, without relying on biometrics or social reputation. The enterprise line continues to use a model where strategies are set by humans within hardware security modules. In September, the company hired Oded Blatman as Chief Information and Security Officer, with a background at Fireblocks; Bloomberg reported that AI-boosted crypto attacks have surged to about $1.4 billion. The company is reportedly exploring a New York listing, with a valuation potentially exceeding $4 billion, and has opened a New York office, hiring a former Circle executive as CFO.
Authorization shifts from software passwords to hardware confirmations that cannot be rewritten by the model. Buyers are users and developers who want to integrate agents into wallets without handing over keys, while sellers are hardware vendors offering trusted screens and strategy engines. The event is driven by Bloomberg interviews and the agent economy roadmap. The beneficiaries are Ledger, which turns "humans in the loop" into a product; the pressured parties are applications that directly hand over keys or passwords to agents, and software wallets that lack a second hardware barrier even after being breached. Probabilistic reasoning can be open, but deterministic authorization must be locked in the chip.
Source: Public Information
ABAB AI Insight
Rogers transitioned from digital officer at LVMH to Chief Experience Officer at Ledger, and in April 2026, his title changed to Chief Human Officer, rewriting a decade of hardware wallet experience into agent governance. The path is to first release a device management suite, then push the agent stack: proposals belong to the model, signatures belong to the screen. The enterprise HSM model of "humans write strategies, machines execute" is replicated in consumer agents. The New York office and listing rumors indicate this narrative is aimed at institutional sales, not just retail in the crypto space.
Capital is shifting from selling cold wallets to selling authorization layers. The agent economy aims to hand over stablecoins, tokenized assets, and login credentials to software, while Ledger keeps the last centimeter on secure chips. The motivation is to prevent agents from being hijacked and directly transferring funds. Resources include trusted displays, strategy engines, and human proof: not collecting biometrics, but proving "there is a unique person who has interacted with this device." The $1.4 billion attack figure is used to demonstrate that software entropy and agent key holding are no longer sufficient.
In comparison to bank USB keys, Apple secure enclaves, and dual approval for enterprise privileged accounts, the industry is moving from "don't hand keys to websites" to "don't hand keys to agents." The phase is redefining alignment with hardware roots: alignment is not about adjusting model weights, but about removing decision-making power from probabilistic systems. Control lies with those who manage the strategy engine, not with those who train larger models.
The essence is a technological replacement of layered pricing power transfer. The execution layer shifts from human confirmation to agent proposals, while the authorization layer shifts from software pop-ups back to hardware. The mechanism is: the more autonomous the model, the more approval must be made into a deterministic circuit that the model cannot touch; whoever controls that screen controls the financial exit in the agent economy. If alignment is only written in prompts, attackers can change the prompts; if written in chip strategies, human presence is required.
ABAB News · Cognitive Laws
The model is responsible for reasoning, the chip is responsible for approval
Run automatically within strategy, must see a human outside strategy
Handing keys to agents equals handing alignment to probability