Corralling AI Risks with Grassroots Contracts

Ehud ShapiroarXiv 2026 · arXiv

TL;DR

Ehud Shapiro proposes "grassroots contracts"—a digital evolution of 19th-century social contract theory—to reclaim the web from centralized AI control. By using a formal mathematical framework to compile social agreements directly into peer-to-peer smartphone apps, the paper demonstrates how to enforce AI accountability and content provenance without relying on central authorities.

What Problem Does This Paper Address?

Modern digital life is dominated by global platforms (like Facebook or X) that are increasingly ceding control to AI. The author identifies three critical risks stemming from this centralization:

  1. AI Control of Platforms: Owners delegate algorithmic control to AI for surveillance and manipulation, potentially losing the ability to reclaim that control.
  2. Unaccountable AI Agents: "Social bots" masquerading as humans can operate on platforms without clear ownership or liability for their actions.
  3. Unattributed Deepfakes: Anonymous AI-generated content can spread rapidly, with producers evading any culpability for harm.

The core issue is that these platforms are monolithic entities governed by "others." Shapiro argues that to mitigate these risks, we need a "Web owned, operated, and governed by the people who use it."

Methods

The paper introduces a rigorous engineering pipeline called the abstraction cascade to turn high-level social goals into functional, secure software.

Figure 1: The abstraction cascade, from a grassroots contract to a running app. Each arrow but the first, an informal translation, represents an implementation, with the reference defining it.

Key components of this method include:

  • Grassroots Contracts: Natural language clauses describing social acts (e.g., "Alice and Bob may jointly decide to become friends").
  • Volition-Guarded Multiagent Atomic Transactions: A mathematical vehicle for specifying platform behavior. A transaction only occurs if the human participants (or their authorized agents) explicitly "will" it.
  • vGLP & GLP: A specialized multiagent logic programming language used to implement these transactions.
  • Peer-to-Peer Architecture: Unlike "decentralized" systems like Bitcoin that still rely on global resources (like a single blockchain), grassroots platforms are constituted solely by the participants' smartphones. There are no central servers to capture.
  • AI-Assisted Coding: Interestingly, the author utilized AI to help code the various layers of the abstraction cascade from the scientific papers describing them.

Datasets

Not discussed in the paper. As a theoretical and systems-oriented paper, it focuses on the mathematical proofs of protocol correctness and the development of prototypes rather than training on or evaluating against standard machine learning datasets. The author mentions tools for "large-scale simulation and testing," but specific benchmarks are not detailed.

Key Insights and Ideas

  • The "Car" Analogy for AI: The paper proposes a breakthrough for AI accountability by treating AI agents like automobiles. Liability is split: the developer is accountable for "manufacturing defects" (bugs/biases), while the principal (the human operator) is accountable for "reckless operation."
  • Provenance over Detection: Rather than trying to build better deepfake detectors (a losing arms race), the system uses digital provenance. Every piece of content is a "digital speech act" signed by its author and every person who forwarded it. If a deepfake appears, someone in the chain is always held accountable.
  • Enforced vs. Attested Clauses: The system distinguishes between rules enforced by the protocol itself (e.g., an AI cannot act without a principal) and rules that leave "non-repudiable evidence" for legal recourse if breached.
  • Resilience to AI Capture: Because the platform has no central server, an AI cannot "take over" the network. To control a grassroots platform, an AI would have to compromise every participant's smartphone individually.

Key Results

The author successfully implemented two primary grassroots contracts:

  1. GAI (Accountable AI Agents): A system where AI agents can only join a social graph under a human principal. Every act by the AI names its principal and developer and carries a cryptographic attestation.
  2. GSNP (Social Network with Provenance): A messaging system where every item carries a list of signatures showing its entire path through the social graph.
Figure 2: The GSG app: Alice asks to be Bob’s friend, and Bob answers Accept or Decline (GSG:1); Alice’s friends after Bob accepts.

The paper provides mathematical proofs (detailed in the Appendices) that these protocols are "volitionally grassroots," meaning they preserve individual agency and allow disjoint groups to coalesce into a larger union by mutual consent.

Limitations and Future Work

  • Competitiveness: The author admits that grassroots smartphone apps are currently not competitive with the polished user experiences of incumbent global platforms.
  • Resource Constraints: As a not-for-profit, common-good project, it lacks the massive engineering resources of Big Tech. The author suggests that using AI to build these "AI-corralling" tools might bridge this gap.
  • Social Adoption: The paper focuses on the technical and mathematical feasibility; the massive social challenge of migrating users from convenient centralized platforms to self-governed ones remains unaddressed.

Summary updated

Abstract

AI's inflection point is a fountain of alarming risks. Digital platforms in particular are vulnerable to three major AI risks, as those in control of global platforms cede control to AI, of the algorithms, of membership and of content: (1) AI control of global platforms, social and economic, used by the majority of humanity, resulting in the majority of humanity becoming vulnerable to AI control. (2) Unaccountable AI agents operating anonymously on social and economic digital platforms, side-by-side with humans, possibly masquerading as humans, serving unknown masters with unknown goals, operating under unknown restraints without accountability or culpability. (3) Unattributed deepfakes generated by anonymous AI, with people unable to distinguish them from genuine content and with their producers evading accountability and culpability. Here, we propose to corral these AI risks with grassroots contracts, a digital incarnation of a two-centuries old notion of Pierre-Joseph Proudhon. We present grassroots contracts for a social graph with accountable AI agents and for a social network with digital provenance, and working smartphone apps that implement the two and show how they corral these major AI risks: (1) AI control of global platforms. (2) Unaccountable AI agents. (3) Unattributed deepfakes.
BibTeX
@misc{shapiro2026corralling,
  author = {Ehud Shapiro},
  title = {{Corralling AI Risks with Grassroots Contracts}},
  year = {2026},
  eprint = {2611.06374},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.06374},
}
Figure