The agrarian and industrial revolutions each transformed civilisation by reorganising what humans and their tools do - but in every case, humans remained the coordinating intelligence that directed the system. The integration of artificial intelligence is different in kind: for the first time, a new autonomous agent is entering the picture, capable of replacing human coordination at a scale and pace without precedent. A capability that proves itself in one setting may be replicated across a sector in months, with no redevelopment between copies, and the terms on which it is adopted lock in just as fast, through self-reinforcing dynamics. Whether those dynamics concentrate gains or distribute prosperity depends on the conditions under which AI enters the economic game. So does whether its costs are carried or pushed out of sight, whether the people and inputs it draws on are recognised, and whether anyone can see how it works or has a say in how it changes.
Shaping those conditions is the central design challenge of our time, and the purpose of the AIC Framework.
The Interaction Layer
In any complex adaptive system, the emergent properties of the whole follow from the rules of interaction between agents, not from the characteristics of any agent taken alone. The same principle applies to the AI transition. The conditions under which AI affects society are shaped at several layers: what a system is designed to deliver, how it works, and what regulation permits and prohibits. All these matter. But from a systems perspective, the highest-leverage layer may be this: the operational arrangements through which AI capabilities are put to work and exchanged between actors. A well-intended, well-designed system, operating within the law, can still produce harmful systemic outcomes depending on the terms governing its use, whether anyone can see how it operates, who captures the gains, and who counts as a stakeholder. The architecture of the deal matters as much as the architecture of the model.
The existing legal and economic instrumentarium at this layer - the logic of value capture, profit maximisation, and cost externalisation that structures commercial relationships - is already straining. Concentration of gains, opacity of operations, ecological costs pushed downstream: these are familiar perils of the pre-AI economy. AI does not even need to introduce any new perils: if it runs the existing ones through systems that replicate at near-zero marginal cost and accelerate without human friction, the toolset that has been producing problematic outcomes at human pace will produce catastrophic ones at machine pace. New interaction patterns at this layer are structurally necessary.
Repeatable Patterns, Systemic Effects
The terms on which AI capabilities are put to work take the form of contracts. The form varies with the parties: a business buying an AI service from a vendor signs a procurement contract; a person using a consumer product accepts terms and conditions; one software system calling another does so under API terms agreed in advance. All these are kinds of contract: an agreement about the terms on which a capability may be used. Each one decides its own narrow specifics: a price, a region, a data right, a notice period. None is asked to weigh how the gains and environmental costs of the AI transition come to be distributed across society, whether anyone outside can see how a system works, or who has standing to contest its terms. Those larger questions are settled in none of these documents - and answered all the same, continuously, as the aggregate that forms when countless private decisions accumulate, each rational on its own terms, none taken with the whole in view. The shape that results belongs to no one: what every contract leaves unsaid allocates as firmly as what it says, and the costs land on whoever was not at the table.
A single contract governs only its own deal - it binds the two parties to it and goes no further. It produces no shared vocabulary and leaves nothing others can read, compare, or build on. System dynamics shift only when a choice becomes repeatable - when a recognisable pattern emerges that actors adopt because it works for them, and that third parties can read, compare, and come to expect.
Creative Commons is the clearest precedent. A small family of standardised terms for sharing knowledge and culture, required by no government, adopted by creators and scholars because it made certain choices easy and legible. Individual decisions, over time, aggregated into something that had not existed before: a commons of openly accessible resources. Open-source licensing achieved the same for code, and now increasingly for AI model weights. Neither was imposed from above; both reshaped entire sectors through bottom-up accumulation alone, and both remain enforceable through existing legal infrastructure.
The AIC Framework applies a similar logic to the conditions under which AI capabilities are contracted into use. Its vehicle is a licence - not in the narrow intellectual-property sense, but in the functional sense that open-source developers already recognise: a set of terms that attaches to a capability and travels with it through the value chain. When a developer chooses an open-source licence, they set conditions on how their code circulates; across thousands of projects, those individual choices built an ecosystem. The AIC licence does the same for AI capabilities - where Creative Commons and open source attach conditions to intellectual property, the AIC Framework attaches conditions to the deployment and use of AI capabilities.
The Framework
The AIC licence does not govern a model, or a company, or "AI" in general. It governs a deployment: a particular configuration of technical, social, and economic arrangements through which an AI capability is put to work, producing a declared deliverable - a product, a service, an offer - and held together by a set of binding documents. A general-purpose AI assistant offered to the public as an online service - its model, its interface, the terms users accept to use it - is a deployment. So is a council using an AI system to triage housing applications, a bank running the same underlying model to score loans, a hospital network monitoring patients, a logistics platform routing freight. The same model can sit inside hundreds of them, each governed separately, because what AIC attaches to is not the technology but each operationally unique arrangement through which it touches the world - and the deliverable the adopter declares delineates each one: every base, boundary, and balance under the profiles follows from it.
The framework is organised around six governance profiles, each independently optional. Each names one flow crossing the boundary of a deployment - something it takes from the world or sends back into it - and asks that the flow be held in balance rather than left to run one way. A deployment draws in data, labour, knowledge, energy and compute, and sends out a service, revenue, emissions, and effects on the people and markets it touches. Each profile picks one of these and brings it into balance; a deployment that activates a profile is choosing to attend to that balance, to count an in or an out its books would otherwise ignore. An adopter selects the profiles relevant to its context, composing a recognisable configuration.
Reciprocity AIC-R ○ ● ○ A commitment to recognising and rewarding the work a deployment runs on but did not pay for: the datasets the model learned from, the people who labelled and curated them, the open-source code, the knowledge its outputs draw on. The adopter routes a share of the value the deployment earns into shared funds for the kinds of contributor that ordinarily go unrewarded, on conservative, dependable terms rather than precise piece-by-piece accounting. Across many adopters, these records accumulate into a shared map of what AI value chains are actually built on. SDG 8 — Decent Work and Economic Growth | Sustainability AIC-S ○ ● ○ A commitment that a deployment's environmental impact is neutral at minimum, and where feasible, regenerative. The adopter measures the deployment's own environmental footprint - energy, emissions, water, hardware - and the environmental effects of what it does in operation, and holds the two in a net balance at source: each deployment accounts for its own impact rather than relying on sector-wide offsetting settled after the fact. If every unit balances where it runs, the total stays bounded however fast the capability replicates - an AI economy that can scale without its environmental draw scaling with it. SDG 12 — Responsible Consumption and Production · SDG 13 — Climate Action |
Openness AIC-O ○ ● ○ A commitment that a deployment's operational logic is legible and open to scrutiny: for a foundation model offered as a service, the model itself - its architecture, training, and evaluation; for a deployment built on someone else's model, the configuration that turns a general capability into this specific operation - the prompts, the tools, the decision rules, the data, the points where a person intervenes. The principle is plain: if it cannot be shown, it cannot be run; and where some detail must stay confidential - the only ground the profile admits is the law's own command - the restriction and its justification are themselves public. Across many adopters, these disclosures accumulate into a shared, comparable record of how AI is actually operated in practice - a commons of operational evidence that anyone can draw on independently of any single provider. | Governance AIC-G ○ ● ○ A commitment that the communities and stakeholders a deployment materially affects have a standing place in the decisions that change how it affects them - a seat in the decision itself. The default schedule is deliberately narrow, two decisions a convened representative body can hold in earnest: the deployment's shutdown, which the body may also initiate by its own motion, and the deployment's AIC commitments, which once Governance is adopted are no longer changed without the body. The adopter constitutes the first body and must state how it genuinely represents the affected; from there the setup is self-governing - its gravest consents public, its composition never closed. Across many adopters, these bodies accumulate into a distributed infrastructure of stakeholder power in the AI economy. SDG 16 — Peace, Justice and Strong Institutions |
Access AIC-A ○ ● ○ A commitment to open access, free of charge, to the exact product or service the deployment offers the market - the offering itself, at its full level, with no degraded variant standing in for it. As AI capability becomes a condition of economic and social participation, access priced by ability to pay widens the divide it inherits - those who can afford the capability gain its advantage and those who cannot fall further behind - so the adopter commits that every private individual holds a free envelope sized for ordinary personal use, with volume the only adjustable constraint. As more deployments carry comparable commitments, a floor of universal availability emerges across the ecosystem, the capability kept open as common infrastructure rather than enclosed as private advantage. | Value AIC-V ○ ● ○ A commitment to directing a minimum of 5% of the deployment's gross revenues into an unconditional basic-income scheme - a periodic, individual cash payment to everyone resident in a defined region, meeting BIEN's criteria: no means-test, no work requirement, no screening. The adopter funds the scheme and names the region - until global poverty is ended, one where a substantial share of residents live below an internationally recognised poverty line - but does not run it: an independent operator holds the distribution, and the adopter never sees a recipient. Across many deployments, these contributions aggregate into a distributed, many-to-many basic-income layer - an income floor that grows with AI's own expansion, aimed at one of the world's oldest unsolved problems. SDG 1 — No Poverty · SDG 10 — Reduced Inequalities |
Across all six, the systemic intervention of the framework is the same: where an AI deployment's costs and consequences would tend to fall on people and environments that are not party to the deal, each profile pulls one of those effects back inside its scope of concern. This can make the AI transition answerable for its effects, and capable of becoming an instrument of the common good.
Composing a Commitment
The AIC profiles are modular: an adopter activates any combination, written as a deployment code - AIC-SOG is Sustainability, Openness, and Governance; AIC-RSOGAV is all six; the code names the profiles a deployment holds for itself. Each commitment is voluntary, but standardised and visible, and each is also directional - a profile binds at one or more of three positions along the value chain: Core, Upstream, and Downstream, written as the profile's propagation pattern: a row of dots - upstream, core, downstream - filled where the duty binds and open where it does not:
○●○ - core only
●●○ - upstream and core
○●● - core and downstream
●●● - the full chain
●○○ - upstream only (et cetera)
Selecting Core applies the commitment to the deployment itself - the one declaring it. Selecting Upstream turns it into a binding requirement on the deployments that supply this one - e.g. the model providers, compute and cloud operators, data and labelling services it depends on. Selecting Downstream turns it into a binding requirement on the deployments built on this one - the operators who take its outputs into their own products and services. Each party sets conditions in its own contracts.
Enforcement
AIC has no separate enforcement apparatus and certifies no one. Adoption is voluntary, but the commitment, once made, binds with the ordinary force of contract - as hiring someone or selling a house is entered freely and then creates real, enforceable consequences. A commitment takes effect through whichever contract or legal channel it is written into, and its force is described by two things: who may enforce it, and how far it reaches.
Who may enforce a commitment depends on how it is declared. It may be the value-chain counterparty that accepted it in a binding agreement, enforcing through that contract's ordinary remedies; the end user, where a profile is declared in terms of service or a published policy and is thereby incorporated into the user contract or actionable under consumer-protection law; a third-party beneficiary the licence expressly names; or a regulator under the applicable law. How far a commitment reaches is a separate question, set not by who can enforce it but by its propagation pattern - the upstream, core, and downstream positions recorded in the dots. A profile whose pattern holds the core alone binds only the adopter's own deployment; one reaching downstream travels to the operators who build on it, who inherit it - the same copyleft logic that carries an open-source obligation to everyone downstream.
What has been published stays checkable: an adopter's records are published cumulatively in the standard's form, anyone may archive them, and deleting or altering the record is itself a breach.
What does the standardising work is clause integrity. A profile name carries a fixed, published meaning, so that "AIC-O" denotes the Openness module exactly as published, the way "CC BY" denotes one defined licence everywhere it appears. An adopter takes on a profile in full or does not claim it - there is no partial or privately edited version wearing the same name - which is what lets a counterparty read a deployment code off a contract and know precisely what it imports. On that fixed-meaning floor the framework's coherence builds: commitments become comparable across deployments through the deployment code and deployment matrix, and checkable against the operational evidence the Openness profile keeps current. This is a different instrument from the certification schemes that dominate AI governance, where a body writes a meta-standard and an accredited auditor issues a seal. The combination of contractual commitment and public declaration works within existing legal systems and requires no new legislation.
AI Commons (AIC) Framework Working Group
The AIC Working Group is currently hosted at CLEA (Center Leo Apostel for Interdisciplinary Studies), Vrije Universiteit Brussel. Its work is directed toward establishing the framework as an open international standard, developed through the kind of multi-stakeholder, publicly-versioned process by which standards earn legitimacy. The group's purpose is to bring the framework from draft stage to a working instrument, combining interdisciplinary governance research with implementation in real AI ecosystems.
The work is organised along three lines:
- Profile architecture review. The overall structure of the framework's profiles - their boundaries, interoperability, and the degree of standardisation they should carry. The objective is to ensure that the system is coherent and implementable across diverse contexts.
- Clause-set development. The working group reviews and finalises the profiles' standard clause sets, bringing each to a first stable version: refining scope and definitions, specifying baseline clauses and adjustable parameters, and maintaining a clear rationale for design choices. This work is taken forward through dedicated expert groups as the profiles require.
- Framework stewardship. Ongoing governance of the framework's components and the processes by which they evolve. The working group serves as the custodian of the framework as a whole, ensuring that changes are coherent, transparent, and aligned with the framework's governance intent.
The framework is a contractual standard - a set of licence terms adopted through ordinary agreements, not a platform or blockchain-based system. It consists of standardised clause sets, profile definitions, and governance templates - designed to work within existing legal and contractual channels. Technical tooling may emerge around it where useful, but no technical layer is constitutive of the framework's operation or centrally controlled.
Members
References
- Bauwens, M., Niaros, V . (2017). Value in the Commons Economy: Developments in Open and Contributory Value Accounting; P2P Foundation. Berlin: Heinrich Böll Foundation. [Link]
- Kostakis, V., Tympas, A. (2025). AI as commons: Why we need community-controlled Artificial Intelligence. Internet Policy Review. [Link]
- Lenartowicz, E.M. (2025). Impact-Oriented Licensing for Artificial Intelligence: A Conceptual Framework for a New Domain of AI Governance. SSRN id=5794362
- Lenartowicz, E.M. (2025). Shaping AI Impacts Through Licensing: Illustrative Scenarios for the Design Space. SSRN id=5835702
- Lenartowicz, E.M. (2025). AI Commons (AIC) Licence Suite: A Modular Framework for Impact-Oriented AI Governance. SSRN id=5848523
- Lenartowicz, E.M. (2026). The AI Commons Framework: Private Ordering as Systemic Co-Regulation Across AI Value Chains. Submitted to the special issue of the Law, AI and Regulation Conference (LAIR 2026), Erasmus University Rotterdam, 11–12 June 2026. SSRN id=6914804