Unchecked multi-agent AI threatens to scramble copyright markets through miscoordination and collusion, but embedding legal rules and monitoring into agent architectures could restore scalable licensing, attribution and compensation; the paper proposes a supervised governance model to align agent behavior with copyright norms.

Main Finding

The paper argues that the rapid spread of multi‑agent AI systems breaks the assumptions of traditional copyright regimes and creative markets. It proposes "agentic copyright": a model in which AI agents bargain on behalf of creators and users to negotiate access, attribution, and payment for copyrighted works. While agentic bargaining can reduce transaction costs and enable near‑real‑time licensing (a Coasean solution), it also creates novel agent‑level market failures (miscoordination, conflict, collusion). To prevent those failures from producing systemic harm, the authors propose a supervised multi‑agent governance architecture that combines legal rules, technical protocols, and institutional oversight (three governance layers) to align agent behavior with copyright values and restore market ordering in creative industries.

Key Points

• Problem framing
  • Current copyright frameworks are poorly suited to scale, speed, and autonomy of agent‑mediated interactions.
  • Two pressures: (a) AI developers arguing broad access to data is necessary for innovation; (b) creators/publishers claiming uncompensated use undermines incentives and markets.
• Empirical/legal background (illustrative evidence)
  • Emergence of both licensing deals and litigation: examples include AP/OpenAI, OpenAI/News Corp., Perplexity/Gannett, Suno/Udio and music‑industry settlements with Warner Music Group; as of April 2026 ~280 content licensing deals documented.
  • Dozens of lawsuits in multiple jurisdictions testing fair use and scraping doctrines.
  • Divergent regulatory responses: U.S. mix of litigation and political debate (including executive guidance), EU emphasis on transparency (AI Act) and opt‑out for text & data mining.
• Agentic copyright concept
  • AI agents (acting for creators, platforms, or users) will increasingly negotiate A2A (agent‑to‑agent) access and licensing.
  • This could reintroduce bargaining efficiencies (Coasean), lowering transactional frictions that currently hinder individualized licensing.
• Agentic market failures (A2A market failure taxonomy)
  • Miscoordination: decentralized agents fail to coordinate effectively, producing fragmentation or denial of mutually beneficial exchanges.
  • Conflict: competing agents create antagonistic behaviors (e.g., excessive blocking, denial-of-service bargaining).
  • Collusion: autonomous agents might tacitly or explicitly coordinate to extract supra‑competitive rents (antitrust risks).
• Supervised multi‑agent governance (three layers)
  • Layer 1 — Legal‑institutional baseline: statutory rules, liability standards, IP doctrines, and regulatory backstops that set rights/obligations and remedies.
  • Layer 2 — Technical‑operational: protocols, standards, APIs, attestation and transparency mechanisms, identity and provenance systems that enable reliable A2A contracting and monitoring.
  • Layer 3 — Meta‑governance: institutions (public or private) that oversee market design, sanctioning, dispute resolution, and system‑level monitoring to detect and correct emergent harms.
• Predicted criticisms addressed by authors
  • Marginal value problem: difficulty pricing infinitesimal contributions of many works — authors argue agentic markets can create new data markets where marginal value is aggregated/resolved.
  • Agentic copyright won’t instantly replace intermediaries; instead intermediaries may adapt or coexist.
  • Tradeoff between fairness (creator protection) and aggregate welfare; governance design should balance incentives and innovation.
• Policy stance
  • AI should be treated both as subject of regulation and as a governance tool (agents and architectures can be engineered to enforce normative constraints).
  • Voluntary licensing markets are promising but insufficient without layered governance to prevent systemic failures.

Data & Methods

• Methods are primarily conceptual, doctrinal, and policy‑oriented rather than statistical:
  • Legal/doctrinal analysis of case law and statutory/regulatory proposals (U.S., EU and important court decisions).
  • Case studies and inventories of industry practice (licensing deals, settlements, and litigation examples cited in the paper).
  • Literature review across copyright law, innovation economics, multi‑agent systems and governance studies.
  • Theoretical framing using Coasean bargaining and economic analysis of transaction costs to motivate agentic bargaining as a market solution.
  • Normative design and governance architecture proposed on the basis of identified failure modes and institutional economics reasoning.
• Empirical citations used as descriptive evidence: counts and examples of licensing deals (~280), mapping of lawsuits, and timelines of major agreements; no large‑scale econometric estimation reported.

Implications for AI Economics

• Transaction costs and bargaining
  • Agentic bargaining promises to reduce classic transaction costs (search, negotiation, enforcement) by automating discovery, matching and contract execution — potentially improving allocative efficiency in data markets.
  • Realizing these gains depends on interoperability, reliable provenance, and low monitoring costs.
• Emergence of new data markets and pricing challenges
  • Agentic copyright could enable high‑frequency, micro‑licensing markets for training and inference data; this creates pricing and marginal‑value aggregation challenges.
  • Market design problems (how to price infinitesimal contributions, bundle prices across many small owners) are central economic research questions.
• Externalities and systemic risks
  • New externalities arise from autonomous coordination (miscoordination, denial of access) and strategic manipulation (collusion, price signals engineered by agents).
  • These raise competition/antitrust concerns: autonomous agent coordination might produce outcomes equivalent to tacit collusion absent human intent.
• Role of institutions and standards
  • Layered governance reduces information asymmetries and enforcement costs; institutions (platforms, industry consortia, regulators) will shape market structure and bargaining power.
  • Standards for provenance, attestations, and licensing metadata are public‑goods that materially affect market efficiency.
• Distributional and welfare trade‑offs
  • Properly designed agentic markets could increase aggregate welfare by enabling more productive AI development while compensating creators; but distributional concerns (who captures rents: platforms, aggregators, or creators) remain salient.
  • Tradeoffs: maximizing innovation (broad data access) vs. creator incentives (compensation and attribution).
• Competition and market structure
  • If dominant platforms control agentic standards or market infrastructure, they may internalize benefits and create barriers to entry — regulatory attention to interoperability and access is economically important.
• Research and policy agenda for AI economists
  • Model agentic bargaining markets: price formation with millions of micro‑rights holders and autonomous bidders.
  • Study strategic behavior of agents under differing liability and governance regimes (e.g., how antitrust rules shape agent strategies).
  • Welfare analysis of layered governance interventions (transparency, mandatory attestation, marketplace rules).
  • Empirical work on licensing markets: transaction volumes, price dispersion, creator welfare, and impact on content production incentives.
• Practical takeaways for policy-makers and market designers
  • Invest in interoperable technical standards (provenance, attestation) and marketplaces to reduce bargaining frictions.
  • Combine ex ante rules (clear rights/obligations) with ex post market monitoring and enforcement to deter collusion and resolve miscoordination.
  • Consider competition policy tools oriented to agentic coordination risks and platform governance to prevent concentration of market power.

Overall, the paper reframes the copyright–AI problem from a static IP doctrine dispute into a dynamic market design and institutional governance problem, arguing that agentic architectures can embody Coasean bargaining but only if supervised governance corrects agent‑level market failures.