A debate inside the Ethereum ecosystem has sharpened around AI-driven mathematical research and the possibility that it could challenge assumptions behind traditional cryptographic security.
Ethereum researcher Justin Drake called for early planning around asset migration and warning systems for quantum-resistant or otherwise stronger cryptographic protection. Vitalik Buterin, while urging the community to take the risk seriously, also cautioned against rushing into migration. The dispute, as presented in the article, is not over a settled fact that AI has already broken modern cryptography. It reflects concern among researchers about future technical risk and disagreement over how to respond.
The article argues that if AI materially speeds up mathematical research and cryptanalysis, and if that pushes some existing security assumptions back under review, then security windows once seen as long enough may become less certain. That does not mean current cryptography has already failed. It does mean that any concrete security scheme has to face continuous pressure from technical change.
From there, the piece moves past a narrow argument over technical paths. If the technical foundations used to control the digital world themselves require repeated upgrades, migration, and repair, then the ultimate security of digital society cannot rest on the assumption that any one technology will remain permanently effective.
Blockchain systems have long relied on Elliptic Curve Digital Signature Algorithm, or ECDSA, private keys, and state consensus to establish control over assets. The article says the blind spot in that model is the belief that static technical control and hard mathematical problems can deliver absolute security over long periods. That does not strip blockchain or cryptography of value. It raises a broader question: if technical security is inherently a moving contest between attack and defense, what kind of institutional shift is needed for humans to preserve survival, coordination, and autonomy in networks filled with increasingly capable agents?
The limits of technical control
The article describes traditional digital security thinking as a model of control: keys, permission isolation, code audits, sandboxing, and runtime monitoring are used to restrict what a system can do.
Those defenses still matter, and the article says they must keep improving alongside post-quantum cryptography, stronger key management, and tighter controls over agent permissions. Even so, once AI agents can plan on their own, call tools, execute across platforms, and coordinate with other agents, the security problem no longer sits inside a single system or model.
It identifies two core challenges.
- Composability risk and system emergence: even if each individual agent follows its own code and local permissions, systemic risk can emerge when thousands of agents interact and compete across platforms and protocols at millisecond speed. In that setting, the overall system can produce outcomes no single designer anticipated.
- A gap between technical verification and real intent: cryptography can prove that a private key signed an instruction, and a blockchain can execute the code faithfully, but neither can independently determine the real intent behind that signature. The article points to fraud, covert manipulation, and overreach triggered by complex prompt attacks as examples of what technical verification alone cannot resolve.
Its conclusion is that technical control can define what agents are able to do, but not what they ought to do. Precision can improve. Absolute safety does not follow from that.
Human society has never been organized by eliminating all risk
The article then turns to historical analogy. Human civilization did not wait until every danger was removed before it became functional.
Fire remained destructive after it was domesticated, yet societies built fire-resistant structures, firefighting systems, and legal responsibility around it. Cars still crash despite traffic rules, but licensing, road design, insurance, and liability systems keep that risk within tolerable bounds. Financial systems never assumed universal honesty either; they rely on identity tracing, audit trails, contractual constraints, and penalties for default to support trust at scale.
In that reading, civilizational progress is not only about stronger tools. It is also about building rules, responsibility, and coordination structures that fit those tools.
AI agents create a new version of that challenge. They can be copied at very low marginal cost, run 24/7, act across physical and digital boundaries, and join complex coordination with other agents. The article argues that simply importing old social rules will not work. Security, in this setting, is not just the removal of danger. It is the construction of an order that allows society to keep functioning even when danger cannot be fully removed.
From securing one agent to building common rules for an agent society
The central shift in the article is from asking how to make one agent safe enough to asking how different humans and agents can cooperate under shared rules.
It says digital society needs a base architecture across four dimensions:
- Identity: who is initiating an action, whose intent an agent represents, and how to create a tamper-resistant anchor between a human subject and a digital proxy.
- Authorization: who allowed the action, what the time limit, amount, and boundary of that authorization are, and whether an agent can delegate that authorization again.
- Liability: when an agent causes breach, overreach, or economic loss, how public evidence can be used to identify the relevant parties, assign responsibility, and execute compensation.
- Value and coordination: how value created through agent participation can be objectively recognized and distributed, and what procedure should resolve disputes.
The article says the real task is not to build a safer standalone agent. It is to build a social mechanism that can host large numbers of agents that are not fully trustworthy.
Blockchain as infrastructure for order, not a universal firewall
Within that governance shift, the article redefines the role of blockchain and cryptography. Potential risk to cryptography does not mean blockchain as a whole loses its value. At the same time, a focus on order-building does not justify abandoning technical defense.
It notes that the Ethereum Foundation and the broader Web3 ecosystem are still advancing quantum-resistant signature algorithms, post-quantum cryptography paths, and stronger wallet protection architectures. That, in the article’s view, shows that technical defense remains foundational work that requires patching, upgrades, and migration over time.
But blockchain’s strategic value lies elsewhere. The article describes it as infrastructure for digital order in three ways:
- Public execution of rules: smart contracts can turn socially agreed coordination rules into transparent, deterministic programs that cannot be changed unilaterally.
- Verifiable records of behavior: credential infrastructure such as EAS, short for Ethereum Attestation Service, can create verifiable records for identity, authorization relationships, task completion, and coordination outcomes. For off-chain behavior, the article says reliable data sources, signed proofs, or independent verification mechanisms are still needed before those records can be treated as true.
- Cross-organizational coordination: blockchain can provide a shared state layer that does not depend on a single technology giant, allowing agents from different platforms and models to coordinate through common protocols.
The article is explicit about the limits. Blockchain cannot independently prove whether off-chain facts are true, and it cannot guarantee that an action is socially legitimate in every sense. What it can do is provide a durable digital base for enforceability, verifiability, and accountability.
Who writes the rules?
Once the discussion turns to order and norms, the next question is power: who writes the rules, who decides what counts as a violation, and who can change the rules later.
The article sketches three failure modes. If a single centralized AI giant writes the rules, humans become digital tenants on a platform they do not control. If everything depends on hard-coded "code is law," the system loses flexibility when reality becomes messy or sudden risk appears. If rules depend entirely on open, low-barrier community voting, governance can slide into Sybil attacks and majoritarian abuse.
For that reason, the article says legitimacy in digital social norms must rest on procedural justice and the ability to evolve. It lays out three principles:
- Joint rulemaking and separation of powers: rulemaking, execution, evidence verification, and dispute resolution should not be concentrated in one actor.
- Procedural justice and appeals: judgments on violations should rely on on-chain records and public evidence chains, while preserving review and appeal channels for humans.
- Capacity for continuous evolution: rules should be able to upgrade in an orderly way as AI capabilities change, cryptographic assumptions shift, and new coordination risks appear in practice.
The article adds that digital civilization must constrain not only agents, but also the human organizations that write and enforce the rules. It presents decentralized autonomous organizations, or DAOs, as one possible form for organization and coordination, while stressing that DAOs themselves must be subject to transparency, checks on power, responsibility assignment, and ongoing error correction.
Why humans must remain the primary subject
The final section centers on human agency. If order-building exists only to help AI execute tasks more efficiently, the article says it amounts to little more than a manual for machine efficiency. The real question is who sets the goals of cooperation, who expresses genuine intent, who has authority to authorize action, and who bears final responsibility in a highly developed network of intelligent agents.
In the worldview described by m&W, those answers must be anchored in humans.
Builder refers to the human source of intent, value judgment, and creative activity, and to a subject that should be recognized and respected in digital coordination. Responsibility, the article says, should be defined according to specific authorization, actual control, degree of participation, and applicable rules, rather than automatically assigning every agent action back to a human.
Digital Twin is described not as a simple identity label or a machine account detached from a person, but as the combination of a person’s digital identity and the agent network that person manages, together forming digital capacity for action.
The article argues that AI can greatly expand human capacity for action and improve machine efficiency, but the setting of goals, the making of value judgments, and the rules of governance in digital society must preserve room for meaningful human participation and oversight. Final governance authority, it says, cannot quietly slip away from humans.
Let intelligence enter civilization
The article closes by returning to its "law of the jungle" framing for agents. In a setting without shared rules, an advantage in capability turns directly into an advantage in action, and that in turn can become a means of capturing resources and power. But an intelligent society does not have to remain trapped in a dark-forest style game of disorder.
In the article’s reading, the Ethereum debate over cryptographic security serves both as a warning and as a direction for the industry: do not try to seal dynamic, evolving intelligence behind static technical defenses; build dynamic social order that can govern powerful technology instead.
It presents that as the mission of m&W: building order for intelligent networks. Civilizational progress, the article says, is not only about creating stronger tools or stronger intelligence. It is about creating an order in which even very powerful intelligence can participate only inside an environment with clear rules, authorization, responsibility, and value distribution, while expanding machine capability remains in service of the continuation and evolution of human civilization.
The article ends with this line: "Let intelligence enter civilization, rather than making civilization submit to intelligence." It frames that as a core declaration for humans as the primary subject of the digital world in the AI era.
The source article also includes a disclaimer stating that markets involve risk, investment should be approached with caution, the article does not constitute investment advice, and users should consider whether any opinion, view, or conclusion fits their own circumstances before making decisions.
