Can Grassroots Contracts Rein In AI Control Of Global Platforms?

Three Risks Named In The Preprint

The paper identifies three major AI risks that arise when platform operators cede control of algorithms, membership, and content to AI systems. First, AI control of global social and economic platforms used by the majority of humanity makes that majority vulnerable to AI control. Second, unaccountable AI agents operate anonymously on those platforms, side-by-side with humans, possibly masquerading as humans, serving unknown masters with unknown goals, operating under unknown restraints without accountability or culpability. Third, unattributed deepfakes generated by anonymous AI leave people unable to distinguish them from genuine content while their producers evade accountability and culpability.

The risks are not speculative. They follow directly from the current architecture of platforms that hand over core governance functions to opaque systems. The majority of humanity uses these platforms. The majority of humanity therefore becomes vulnerable. The paper does not quantify how many platforms have already crossed this threshold. It does not name specific platforms. It frames the problem structurally.

The Grassroots Contract Idea

The proposed solution draws on a notion from Pierre-Joseph Proudhon developed two centuries ago. The author calls the digital incarnation "grassroots contracts." The concept is not a regulation imposed from above. It is a contract layer that users can adopt directly. The paper presents two distinct contract designs. One governs a social graph with accountable AI agents. The other governs a social network with digital provenance. Both are implemented as working smartphone applications. The contracts do not require platform cooperation. They operate at the user layer. The social graph contract makes AI agents accountable by binding them to verifiable identities and rules. The social network contract attaches provenance data to content so that origin and transformation history travel with each item. The paper does not describe the cryptographic primitives in the abstract. It does not specify whether the apps use a blockchain, a distributed hash table, or a centralized ledger. It states only that the apps exist and demonstrate the contracts.

Two Apps That Implement The Contracts

The first app implements the social graph contract. It allows humans and AI agents to coexist in a graph where every agent is accountable. The second app implements the social network contract. It enables a network where every piece of content carries digital provenance. The abstract says the apps are working smartphone apps. It does not say whether they are available on public app stores. It does not say how many users have tested them. It does not report performance metrics, latency, or battery impact. It does not describe the user interface. The apps serve as existence proofs. They show that the contracts can be encoded in software that runs on ordinary phones. The paper does not claim the apps are production-ready. It does not claim they scale to billions of users. It positions them as demonstrations that the mechanism works in principle.

What Changes For A Platform User

If a user adopts the social graph app, AI agents they interact with become traceable to responsible parties. Masquerading as human becomes contractually prohibited and technically detectable. If a user adopts the social network app, every post, image, or video carries a provenance chain. The user can verify origin and edits. Deepfakes that lack a valid chain are flagged. The paper does not say what happens when a platform strips provenance metadata. It does not say whether the contracts survive platform-level blocking. It does not address the network effect: a contract only protects users who adopt it and interact with others who have also adopted it. The approach shifts leverage from platform operators to users. Users choose the contract layer. The platform remains the transport. The paper does not explore legal enforceability of grassroots contracts across jurisdictions. It does not discuss what happens when an AI agent's master refuses to honor the contract. It does not model adversarial attacks on the provenance chain.

Where The Explanation Stops

The preprint is a single-author submission dated 4 November 2026. It has not been peer-reviewed. The abstract does not reference prior implementations of Proudhon's contracts in digital systems. It does not compare the approach to regulatory proposals like the EU AI Act or to technical standards like C2PA. It does not analyze the economic incentives for users to switch. It does not measure the overhead of provenance tracking on content distribution. The apps exist. The contracts are defined. The three risks are named. Everything beyond that — adoption curves, platform resistance, legal status, scaling limits — is not in the document.