Altman vs. Amodei: In the Age of Superintelligence, Will Humans Be in Control—or Kept as Pets?
Altman vs. Amodei: In the Age of Superintelligence, Will Humans Be in Control—or Kept as Pets?
What happens if artificial intelligence becomes better than people at almost every economically valuable task? The answer depends not only on how much wealth AI can create, but also on who controls access to it.
OpenAI CEO Sam Altman has argued that humans should retain control over their future, even if doing so means limiting the development of superintelligent AI. His warning is aimed at a particular possibility: a society in which an AI system judges human behavior and determines what rewards people receive.
The idea he challenged appears in a 2024 essay by Anthropic CEO Dario Amodei. But their apparent disagreement needs context. Amodei raised AI-directed distribution as one possible economic arrangement, not as a settled recommendation, and acknowledged that such arrangements could lead to exploitation or dystopian outcomes.
For crypto users, the debate raises a practical question: If intelligent systems become central to the economy, can people still own, move, and use their assets without asking those systems for permission?
The Economic Question Behind the AI Debate
In Machines of Loving Grace, Amodei considers a future in which AI performs most work more effectively and cheaply than humans. If that happens, wages may no longer be a reliable way to distribute the benefits of economic production.
One possible response is universal basic income: people receive resources regardless of whether they have paid employment. Another, more troubling possibility is that AI-generated wealth could be distributed according to assessments of which human activities deserve support.
The difference is significant. A system that guarantees people resources under rules they can understand and contest is not the same as one that makes access conditional on an opaque assessment of their behavior.
Altman’s objection focuses on that loss of agency. The issue is not simply whether an AI system could provide material comfort. It is whether people would retain meaningful authority over the choices, institutions, and resources that shape their lives.
Why Ownership Matters in an AI-Driven Economy
As AI agents become more capable of conducting research, making purchases, and managing workflows, they may also take on a larger role in financial activity. A business could authorize an agent to pay for cloud computing, for example, or an individual could use one to help manage routine transactions.
That convenience creates a distinction: an AI agent can act for a user without owning the user’s assets or holding unrestricted authority over them.
Crypto wallets and smart contracts offer ways to express that distinction in code:
- A user can hold assets in an account they control rather than leaving them entirely with a service provider.
- An application can request permission to interact with assets, though the user must still assess what that permission allows.
- Smart contracts can enforce predefined conditions, such as spending limits or multiple approvals.
- Public blockchains can make certain transactions and rules independently observable.
None of these features guarantees a fair outcome. They can, however, give users tools to separate assistance from authority—a useful principle when software is making decisions at machine speed.
AI Agents Need Permissions, Not Unlimited Access
For an AI agent to be useful with crypto, it may need the ability to initiate transactions. Giving it access to an unrestricted private key, however, could expose the entire wallet if the agent is compromised, manipulated, or simply makes a mistake.
A safer design starts with limiting what the agent can do. Depending on the wallet and infrastructure, that could mean:
- Setting spending caps for a specific period or task.
- Restricting destinations to approved addresses or contracts.
- Requiring human approval for larger or unusual transactions.
- Separating agent funds from long-term holdings.
- Revoking permissions when an agent or application is no longer trusted.
Account abstraction is one area of Ethereum development that can support more flexible account behavior. The ERC-4337 standard describes an approach to smart contract accounts with customizable validation logic. Such designs can help developers build more tailored authorization flows, although the protections available depend on the wallet’s implementation and the rules a user actually enables.
There is no universal safety switch. A spending limit is only as useful as its configuration, and a malicious contract or compromised interface can still mislead a user into granting dangerous permissions.
Transparency Is Valuable, but It Is Not the Same as Control
Blockchain technology is sometimes presented as an automatic answer to centralized power. The reality is more complicated.
A transaction may be visible on-chain while the decision that caused it remains opaque. An AI system might use a proprietary model, private data, or undisclosed criteria to decide who receives a payment. Recording that payment on a public ledger does not reveal whether the decision was reasonable.
Likewise, a smart contract can execute rules consistently without making those rules fair. If a small group can change the contract, control the data it relies on, or block access through a dominant interface, users may have less autonomy than the on-chain record suggests.
For any AI-linked financial system, useful questions include:
- Who can change the rules?
- What data influences a decision?
- Can users understand and challenge an outcome?
- Can people withdraw their assets or switch providers?
- What happens if the AI system or its operator fails?
These questions matter as much as transaction speed or automation. Human control requires both technical access and institutions that make that access meaningful.
Where Self-Custody Fits
Self-custody gives users direct responsibility for the keys that authorize transactions. In an economy increasingly mediated by automated services, that can provide an important boundary: an AI assistant may recommend an action, while the asset owner retains the ability to decide whether to sign it.
A hardware wallet such as OneKey can support that separation by keeping private keys on a dedicated device and requiring transaction approval through the wallet’s signing flow. It does not judge whether an AI recommendation is sound, and it cannot protect a user who approves a malicious transaction after being deceived. Users still need to verify transaction details and safeguard their recovery information.
For everyday experimentation with AI agents, keeping limited funds in a separate wallet and reserving long-term assets for more deliberate approvals can reduce exposure. The right setup depends on the assets, applications, and risks involved.
The Goal Is Agency, Not Just Abundance
The Altman–Amodei debate is not a simple choice between welcoming AI and rejecting it. Both the economic promise and the risks deserve serious attention. The harder question is how to ensure that growing machine capability expands human choices rather than quietly replacing them.
Crypto cannot resolve that question on its own. But its focus on ownership, portable assets, and explicit authorization can contribute to the design of systems in which AI acts as a tool—not the final authority over what people may do with their resources.
