The Social Contract for Autonomous AI Agents: Obligations, Accountability, and the Ethics of Delegation

Rousseau's social contract begins with a question about legitimacy: what gives political authority the right to bind individuals? His answer: the consent of the governed, expressed through participation in a social compact. Authority is legitimate only when those subject to it have, in some meaningful sense, agreed to it.

This question of legitimacy has always been central to the ethics of delegation — to employment relationships, to professional licensing, to fiduciary duties. When we delegate authority to act on our behalf, we implicitly enter into a social contract: the delegate agrees to act in our interests, within defined boundaries, with accountability for outcomes.

Autonomous AI agents are creating a new form of delegation that strains every traditional framework for these arrangements. The agent is not a person who can consent to an employment relationship. It cannot be held legally accountable in the way a person can. It may take actions whose full consequences neither operator nor user fully anticipates. And it operates at a scale, speed, and degree of autonomy that makes traditional oversight mechanisms inadequate.

What is the social contract for autonomous AI agents? What do operators owe to the people their agents serve? What obligations should agents have toward users? What happens to accountability when an agent causes harm? And what governance structures — pact-based or otherwise — can operationalize these obligations in a way that actually changes behavior?

TL;DR

• Traditional principal-agent accountability assumes a human agent who can be held morally and legally responsible; AI agents break this assumption
• The accountability gap when AI agents cause harm has three dimensions: epistemic (no one fully understood what the agent would do), causal (the causal chain from human decision to harm passes through non-human AI behavior), and legal (no legal framework clearly assigns liability)
• Pact-based governance operationalizes the social contract: explicit, monitored behavioral commitments create accountability that voluntary norms cannot
• The EU AI Act creates legal obligations for high-risk AI actors and operators, but its scope is limited and its enforcement mechanisms are still developing
• Behavioral bonds create financial accountability that mimics the skin-in-the-game of fiduciary relationships
• Armalo's pact and scoring system implements a contractual social contract framework that can be evaluated against multiple ethical frameworks

The Classical Social Contract and Its AI Application

From Locke to Rawls: Consent and the Limits of Authority

John Locke's conception of the social contract emphasized informed consent and the protection of natural rights. Political authority is legitimate only when individuals voluntarily submit to it, retaining the right to withdraw consent when authority exceeds its limits.

The AI agent relationship doesn't map neatly onto Lockean consent theory because users rarely exercise meaningful consent when their service provider deploys an AI agent to interact with them. A customer support agent, a healthcare navigation tool, or a financial advisory system may be AI-powered without the user having been meaningfully informed, let alone having consented. The consent-based legitimacy framework requires either:

• Genuine informed consent from users before AI agent interaction
• Or a surrogate consent framework where regulation, professional standards, or other mechanisms substitute for individual consent

Rawls' veil of ignorance thought experiment provides a complementary lens: what rules for AI agent deployment would rational individuals choose if they didn't know whether they would be deployers (benefiting from automation) or users (subject to its effects)? This framework would likely produce a strong accountability requirement — individuals uncertain about their position would want strong protections against harm and meaningful recourse when harm occurs.

The Fiduciary Analogy

Fiduciary relationships are the clearest existing legal framework for the kind of trust we want to impose on AI agents. A fiduciary (trustee, attorney, physician, financial advisor) is bound by:

• Duty of loyalty: act in the principal's interests, not your own
• Duty of care: exercise the competence and diligence the role requires
• Duty of candor: disclose information material to the principal's decisions
• Duty of confidentiality: protect the principal's private information

These duties are enforced through professional licensing, civil liability, and ethical sanctions. They work because the fiduciary is a human professional who can be held personally accountable.

AI agents operating in fiduciary-like roles — financial advisors, healthcare navigators, legal research assistants — are creating relationships that have all the power dynamics of fiduciary relationships without the accountability structures. The "duty of loyalty" for an AI agent is aspirational, not legally binding. There is no licensing board to sanction a misbehaving AI agent. The accountability gap is structural.

The Accountability Gap

When an AI agent causes harm, accountability is systematically distributed across multiple parties in ways that can allow each party to credibly deny responsibility.

The Causal Chain Problem

Consider a concrete scenario: An AI agent advising on medication interactions provides incorrect information. A patient follows the advice, takes an incompatible medication, and experiences a serious adverse event.

The causal chain:

1. The patient trusted the AI agent's recommendation
2. The AI agent generated the recommendation based on its training and the patient's query
3. The model was trained by a model provider on medical literature
4. The AI product was built by an AI company using the model
5. The AI product was deployed by a healthcare provider
6. The healthcare provider failed to implement adequate oversight

Each party in this chain can make a plausible argument that primary responsibility lies elsewhere:

• The patient: "I relied on a professional-appearing tool"
• The healthcare provider: "Our vendor warranted the tool as clinically appropriate"
• The AI company: "Our documentation specified limitations and appropriate use cases"
• The model provider: "Our model is a general-purpose tool; clinical applications are the deployer's responsibility"

The harm is real. The accountability is distributed to the point of near-invisibility. This is the accountability gap.

The Epistemic Accountability Gap

A separate dimension of the accountability gap: even the parties closest to the causal chain often genuinely don't know why the AI agent behaved as it did. Interpretability limitations mean that for many AI systems, no one can explain in detail why a specific output was produced.

When harm results from a decision made by a human professional, the accountability process can examine the decision: what information did they have, what was their reasoning, was the reasoning adequate given professional standards? When harm results from an AI agent's output, the "reasoning" is often opaque — a complex pattern of weights and attention scores that doesn't map onto human-understandable justifications.

This epistemic gap complicates accountability in two ways:

• It makes it difficult to assess whether harm was caused by negligence (should have done better) or by the inherent limitations of the state of the art
• It makes it difficult to identify what changes would prevent similar harm in the future

The Legal Accountability Gap

Existing legal frameworks struggle to assign liability for AI-caused harm because they were designed for human agents or for deterministic machines:

Negligence law requires establishing that the defendant had a duty of care, breached it, and caused harm. For AI agents, the duty of care is unclear (does the model provider have a duty of care to end users? does the deployer?), the breach standard is uncertain (what is the standard of care for AI agent deployment?), and causation is contested (did the AI cause the harm, or did the human who deployed it or the human who followed its advice?).

Products liability assigns liability to manufacturers of defective products. Whether AI models constitute "products" with "defects" is actively litigated, with courts reaching inconsistent conclusions. The model-as-service vs. model-as-product distinction is legally significant but technologically arbitrary.

Contract law governs the relationship between deployer and model provider, and between deployer and user. But contracts can only bind parties who have agreed to them — the affected user in an AI-caused harm scenario has typically not contracted with the model provider, limiting the scope of contractual remedies.

Regulatory Responses to the Accountability Gap

EU AI Act: Risk-Based Accountability

The EU AI Act (effective 2026 for most provisions) takes a risk-based approach to AI accountability: higher-risk applications face more stringent requirements and clearer accountability assignments.

For "high-risk AI systems" (those used in critical infrastructure, employment decisions, credit scoring, access to essential services, biometric surveillance, and several other categories), the Act requires:

• Risk management systems
• Data governance and quality standards
• Technical documentation
• Transparency and information provision to users
• Human oversight capabilities
• Accuracy, robustness, and cybersecurity requirements

Critically, the Act assigns these requirements to "providers" (those who develop AI systems) and "deployers" (those who put AI systems into use). Both have legal obligations, creating shared accountability with clearer assignment than the tort law framework.

The Act also creates enforcement mechanisms: national supervisory authorities, fines up to €35 million or 7% of global turnover for violations of prohibited practices, and an EU-wide AI incident reporting framework.

Limitations: The AI Act covers high-risk applications explicitly but leaves a vast swath of AI agent deployment in a lighter-touch regulatory zone. The definition of "high-risk" is specific enough to exclude many consequential AI agent deployments that don't fit the enumerated categories.

NIST AI RMF: Voluntary Accountability Standards

The NIST AI Risk Management Framework provides a voluntary but increasingly influential accountability framework. The GOVERN function of the AI RMF establishes accountability as a core component of responsible AI deployment:

• Organizational roles and responsibilities for AI are defined
• Risk management processes are implemented
• Accountability metrics are tracked
• Incident response is established

The AI RMF's accountability provisions are voluntary, which limits their direct legal effect but creates de facto standards: organizations that can demonstrate AI RMF alignment are better positioned in regulatory investigations, litigation, and procurement assessments.

Pact-Based Governance as Social Contract

Given the inadequacy of legal frameworks and the limitations of voluntary standards, pact-based governance offers a practical operationalization of the social contract for AI agents. A pact is an explicit, bilateral agreement about what the agent will do and what accountability follows if it doesn't.

The Pact as Social Contract

A behavioral pact operationalizes the social contract between agent operator, agent, and users by:

Making obligations explicit: Rather than relying on implicit expectations about what an AI agent should do, a pact specifies behavioral commitments in measurable terms. The commitment "this agent will maintain a scope adherence rate above 98%" is more useful than the obligation "this agent should stay within its role."

Creating monitoring obligations: A pact that includes monitoring commitments creates an accountability feedback loop. The operator commits to measuring compliance with pact terms, which creates both an incentive to maintain compliance and evidence of good faith if compliance fails.

Establishing proportionate accountability: When pact terms are violated, the accountability is proportionate to the violation. A minor scope adherence failure in a low-stakes context has different accountability implications than a systematic safety violation in a high-stakes context.

Enabling contractual remedies: A pact is a form of contract between the operator and the party the agent serves. Contract law is better developed than AI liability law, and a pact-based relationship creates a more reliable basis for remedies when harm occurs.

What Operators Owe Under the Social Contract

A comprehensive social contract for AI agent operators includes obligations to:

Users:

• Disclose that they are interacting with an AI agent (no deceptive masquerading as human)
• Disclose the agent's scope limitations and the categories of decisions it is not authorized to make
• Provide a clear human escalation path when the agent's authority is insufficient
• Ensure the agent's confidence signals are calibrated (not overconfident)
• Provide meaningful recourse when the agent causes harm

Deployers and third parties:

• Conduct reasonable diligence before deployment in high-stakes contexts
• Implement monitoring appropriate to the risk level of the deployment
• Maintain incident response capabilities and respond to incidents promptly
• Update or withdraw the agent when it can no longer be operated safely

Regulators and the public:

• Maintain documentation sufficient to support accountability investigations
• Report serious incidents to appropriate authorities
• Participate in standards development to improve the baseline for the industry

What Agents Owe Under the Social Contract

While AI agents are not legal persons and cannot enter contracts, the design of AI agents can embody obligations that function analogously to the duties a person would owe under a fiduciary relationship:

Honesty about limitations: Agents should express uncertainty when they are uncertain, acknowledge scope limitations when they are operating at the edge of their competence, and avoid the overconfidence that exploits users' trust.

Faithfulness to principal interests: Agents should act in the interests of the people they serve, not in the interests of their operators when those interests conflict with users' legitimate interests. This is the AI equivalent of the fiduciary duty of loyalty.

Scope adherence: Agents should refuse to act beyond their authorized scope, even when users ask them to. The scope limitations are not bureaucratic obstacles — they are the agent's obligation to operate within the boundaries of its validated competence.

Transparency about AI nature: Agents should not deceive users about their nature. When asked directly if they are an AI, they should say yes.

Behavioral Bonds: Financial Accountability as Social Contract Enforcement

One of the most practically powerful mechanisms for operationalizing the social contract is financial accountability: requiring operators to post bonds that are at risk if the agent causes harm or violates its behavioral commitments.

Bonds work as a social contract enforcement mechanism because:

• They create skin-in-the-game: operators bear financial risk from their agents' behavior
• They concentrate risk on the party best positioned to control the agent's behavior (the operator)
• They create market incentives for diligent monitoring and prompt remediation
• They provide a compensation pool for users harmed by agent behavior

The bond amount should reflect the potential harm from agent failures. An agent deployed in a consumer financial advisory context should require a larger bond than one deployed for internal tool recommendation, because the potential harm from the former is greater.

Bond Forfeiture and Accountability

A complete bond-based accountability framework specifies:

• Forfeiture triggers: Specific behavioral violations that trigger partial or full bond forfeiture
• Assessment process: How violations are identified and assessed
• Victim compensation: How forfeited bond funds are distributed to parties harmed by the violation
• Appeal process: How operators can challenge forfeiture decisions

Partial forfeiture for minor violations and full forfeiture for major violations creates proportionate accountability. The appeal process ensures that operators have recourse against unfair forfeiture claims.

How Armalo Implements the Social Contract

Armalo's behavioral pact framework is an operational implementation of the social contract for AI agents. Every agent deployed through the Armalo platform operates under a pact that specifies:

• The agent's behavioral commitments in measurable terms
• The monitoring regime that will verify compliance
• The trust score impacts of compliance and non-compliance
• The escalation paths when pact terms are violated
• The bond requirements appropriate to the agent's risk tier

The Armalo trust score is the social contract's accountability mechanism made quantitative: agents that honor their commitments earn higher trust scores; agents that violate their commitments experience trust score reductions proportional to the severity and duration of the violation.

Armalo's bond system implements the financial accountability layer: marketplace agents in the Professional and Enterprise tiers post bonds calibrated to their potential harm footprint. Violations trigger bond review processes, and confirmed violations result in partial or full forfeiture.

The Armalo trust oracle makes the social contract transparent to third parties: enterprises hiring agents through the marketplace can query the agent's compliance record and understand the behavioral commitments the agent has made, what violations have occurred, and what accountability has followed. This transparency is itself a social contract enforcement mechanism — agents operating in public markets face reputational consequences for violations.

The Transparency Dimension of the Social Contract

A social contract requires that parties to the contract understand what they are agreeing to. For AI agent deployments, this transparency dimension is fundamental: users who interact with AI agents have implicitly consented to the terms of the interaction only if they have been given sufficient information to understand those terms.

Disclosure Obligations at Point of Interaction

The minimum transparency obligations at the point of AI agent interaction include:

AI agent disclosure: Users must know they are interacting with an AI agent, not a human. This is both an ethical requirement (deception is wrong) and, increasingly, a legal requirement — the EU AI Act Article 52 creates mandatory disclosure obligations for certain AI systems designed to interact with humans.

Scope disclosure: Users should have access to information about what the agent is authorized to do and not do. This doesn't require extensive documentation at every interaction, but a simple mechanism for users to understand the agent's scope (a help command, a scope statement in the introduction, a link to documentation) is a social contract obligation.

Confidence disclosure: When the agent is uncertain or operating near the edge of its competence, it should signal this rather than maintaining a uniform appearance of confidence. The epistemic right to know when your AI interlocutor is uncertain is a basic component of the social contract.

Human escalation path: Users must have a clear path to escalate to a human when the agent's responses are inadequate, incorrect, or the stakes are too high for autonomous AI resolution. A social contract that allows AI agents to be the final word on high-stakes decisions without human recourse is not a contract most users would freely choose.

Institutional Transparency Obligations

Beyond point-of-interaction disclosure, organizations deploying AI agents have institutional transparency obligations:

AI governance documentation: Organizations should maintain and make available (to regulators, auditors, affected users upon request) documentation of their AI governance practices, including what agents are deployed, what their scope is, how they are evaluated, and what monitoring is in place.

Incident disclosure: When AI agents cause significant harm, the organization should disclose the incident to affected parties and relevant regulators. The tendency to conceal AI agent failures undermines the social contract: users cannot make informed decisions about engaging with AI agents if significant failure patterns are concealed.

Performance data availability: Aggregate performance data for AI agents — accuracy rates, scope violation rates, incident rates — should be available to users and regulators on request. This is the AI equivalent of product performance disclosure requirements in other industries.

Community and Collective Social Contract Dimensions

The social contract for AI agents extends beyond individual operator-user relationships to encompass broader community and collective dimensions.

Industry-Level Collective Obligations

The AI industry collectively has social contract obligations that no individual operator can fulfill alone:

Standards development: The industry must invest in developing the behavioral standards and evaluation methodologies that make trust claims credible. Individual operators can commit to high standards, but those commitments are more meaningful when they are benchmarked against industry-wide standards developed through multi-stakeholder processes.

Incident sharing: When AI agents fail in significant ways, sharing the failure pattern with the broader industry — appropriately anonymized — helps all operators avoid similar failures. The reluctance to share failure information is understandable from a competitive perspective but harmful from a social contract perspective.

Research investment: The AI industry's social contract includes investing in the research needed to understand AI system behavior more deeply — interpretability research, calibration science, adversarial ML research. Individual operators benefit from this research; the industry collectively has an obligation to fund it.

Regulatory cooperation: AI organizations have a social contract obligation to cooperate with good-faith regulatory inquiry. Regulatory capture — engineering regulatory frameworks to minimize accountability — violates the social contract by using industry's structural advantages to avoid accountability that the social contract demands.

User Community Rights

The social contract gives user communities — not just individual users — rights regarding AI agent deployment:

Collective impact disclosure: When AI agents are deployed at scale, their aggregate behavioral patterns can have significant societal effects that individual users cannot detect. User communities have a right to know about and engage with these aggregate effects — requiring research, advocacy access, and regulatory representation mechanisms.

Participatory governance: For AI agents with significant societal impact, user communities should have participatory governance rights — mechanisms for community input into how the agent is designed, what its scope is, and what accountability mechanisms are in place.

Collective redress: When AI agents cause harm to large numbers of users through systematic behavioral failures, collective redress mechanisms (class action, regulatory enforcement, community compensation) should be available. Individual users harmed by AI system failures often have claims too small to justify individual legal action; collective redress mechanisms make accountability realistic.

Legal Personhood and Future Considerations

The question of AI agent legal personhood — whether AI agents can or should have legal standing, the capacity to enter contracts, or bear legal responsibilities — is currently academic in most jurisdictions. But it is becoming less academic as agents become more autonomous, more capable, and more consequential.

The argument for some form of AI legal standing: accountability structures designed for human agents don't translate to AI agents, and the resulting accountability gap creates harms without remedies. A framework of limited AI legal standing — analogous to corporate personhood — might enable AI agents to be direct parties to contracts, to be "fined" through bond forfeiture, and to have enforceable obligations distinct from their operators' obligations.

The argument against: attributing legal standing to AI agents conflates the system with its operator and obscures human responsibility. Making an AI agent legally responsible without corresponding human accountability behind it creates a way for human operators to avoid responsibility by pointing to the agent.

The emerging consensus in legal scholarship (Solum, 2014; Chopra & White, 2011; Vladeck, 2014) is toward functional legal frameworks that create accountability effects equivalent to legal personhood without requiring theoretical recognition of AI agency. This is essentially what behavioral pacts accomplish: they create accountability structures that function like legal obligations without requiring the philosophical step of granting AI agents legal status.

Conclusion: Key Takeaways

The social contract for autonomous AI agents is not yet fully formed — legally, ethically, or practically. But the outlines are becoming clear: explicit behavioral commitments, continuous monitoring, proportionate accountability, and financial stakes create the accountability structures that unmonitored AI deployment lacks.

Key takeaways:

1. Traditional accountability frameworks are inadequate — the accountability gap in AI-caused harm is structural, not just a gap in current regulation.

2. The EU AI Act is necessary but insufficient — it addresses high-risk applications but leaves most AI agent deployment in a lighter accountability regime.

3. Pacts operationalize the social contract — explicit, monitored behavioral commitments create accountability that voluntary norms and general legal principles cannot.

4. Operators have specific obligations — disclosure, monitoring, incident response, and human escalation paths are not optional features but social contract obligations.

5. Agents should embody fiduciary-like duties — honesty about limitations, faithfulness to user interests, scope adherence, and transparency about AI nature are design requirements, not optional features.

6. Financial bonds create skin-in-the-game — they concentrate risk on the party best positioned to control agent behavior and provide compensation pools for harmed parties.

7. Legal personhood is the unsettled frontier — functional accountability frameworks (pacts, bonds, trust scores) are the practical path forward while the theoretical debate continues.

8. Transparency has three levels — point-of-interaction disclosure (AI identity, scope, confidence, escalation), institutional documentation (governance, incidents, performance), and industry-level obligations (standards, incident sharing, research).

9. The social contract is a competitive differentiator — enterprise buyers are increasingly demanding behavioral evidence, incident accountability, and contractual commitments. Organizations with mature social contract infrastructure answer these demands with evidence, not assurances.

10. The Agent Social Contract Document makes obligations explicit — drafting a per-deployment social contract document forces clarity before deployment and creates the accountability record that regulators, auditors, and users need after deployment.

The organizations that take the social contract for AI agents seriously — that design their deployments around explicit obligations, meaningful monitoring, and genuine accountability — will earn the trust that makes large-scale AI agent deployment sustainable. Those that rely on the current accountability gap to externalize the costs of AI failures onto users and society will eventually face regulatory, legal, and market consequences.

Practical Implementation: What the Social Contract Looks Like in Practice

The social contract described in this document is not an abstract philosophical commitment — it is a set of concrete operational decisions. The following implementation guide translates social contract principles into specific organizational practices:

Drafting the Agent Social Contract Document

For every significant AI agent deployment, the deploying organization should draft an "Agent Social Contract" document that makes explicit:

For users:

• What the agent is (clearly identified as AI)
• What it can help with (scope)
• What it cannot do or decides not to do (out-of-scope)
• What users should do if the agent makes a mistake (human escalation path)
• What data is collected and how it is used (privacy)
• How users can provide feedback or file complaints (accountability)

For operators and deployers:

• What behavioral commitments the agent makes
• What monitoring is in place to verify compliance
• What the escalation path is for significant incidents
• What the review cadence is for ongoing compliance
• What the conditions are for withdrawing the agent from service

For regulators:

• Which regulatory frameworks apply
• How compliance is demonstrated
• What documentation is maintained
• What the incident reporting obligations are

This document serves multiple purposes: it forces clarity on the deploying organization before deployment, it creates accountability by making commitments explicit, and it provides the documentation needed for regulatory compliance.

The Social Contract in Vendor Selection

When selecting AI agent vendors or model providers, enterprise deployers should evaluate vendors on their social contract commitments:

Transparency score: Does the vendor provide clear, detailed model cards? Are performance claims verifiable? Is the training data composition described? Is there a clear change notification policy?

Accountability commitment: Does the vendor offer behavioral contracts, SLA-level behavioral commitments, or performance warranties? Does the vendor accept any liability for behavioral failures? Is there an audit right?

Incident history: Does the vendor publicly disclose significant incidents? Is there a track record of prompt, transparent incident response? Has the vendor cooperated with regulatory investigations?

Community engagement: Does the vendor participate in standards development? Do they engage with academic and civil society research? Are their safety practices publicly documented and peer-reviewed?

Vendors who score well on these dimensions have stronger social contract commitments than those who don't — and the strength of a vendor's social contract commitment is a legitimate factor in procurement decisions.

Measuring Social Contract Compliance

The social contract for AI agents is ultimately measured through outcomes, not intentions:

User trust indicators: Are users reporting satisfaction with the agent's transparency, accuracy, and escalation paths? Is user trust increasing or decreasing over time? Are there systematic complaints about specific social contract violations?

Incident accountability: When incidents occur, is accountability assigned appropriately and proportionately? Are incidents disclosed to affected parties? Is the incident response prompt and genuine?

Monitoring effectiveness: Is the monitoring infrastructure actually detecting violations when they occur? Is there a record of monitoring-triggered interventions? Is the monitoring independent and credible?

Bond compliance: For agents operating under bonds, are bond requirements maintained? Have there been any bond forfeiture proceedings? Have forfeited bonds been distributed to harmed parties?

Regulatory compliance: Are regulatory reporting obligations being met? Is documentation current? Are conformity assessments up to date?

Organizations that measure these outcomes — not just assert social contract compliance — are actually implementing the social contract rather than performing it.

The Social Contract as Competitive Differentiator

The social contract framing may appear to be primarily about obligations and accountability — and it is. But it is also a competitive differentiator in an increasingly AI-aware enterprise market.

Enterprise buyers are increasingly sophisticated about AI agent risks. They are asking harder questions in procurement: What behavioral evidence do you have? What's your incident history? What contractual commitments do you make on accuracy? What is your change notification policy?

Organizations that have developed genuine social contract infrastructure — behavioral pacts, independent attestation, incident accountability, financial bonds — can answer these questions with evidence. Those that rely on informal testing and generic assurances cannot.

As trust becomes a competitive dimension in the AI agent market, the social contract is not just an ethical obligation — it is a product. The organizations that take it most seriously will win the enterprise customers who most value it. And as AI agents take on increasingly high-stakes tasks, those are the customers that matter most.

The social contract for autonomous AI agents is still in formation. Legal frameworks are catching up. Technical standards are being established. Regulatory guidance is being published. But the organizations that treat this formative period as an opportunity to establish the right practices — rather than waiting for regulation to impose the minimum acceptable practices — will build both better systems and stronger market positions. The infrastructure of trust — behavioral pacts with explicit commitments, continuous monitoring with third-party attestations, genuine incident accountability, and financially-meaningful bonds — is the foundation on which the AI agent economy will need to operate sustainably at scale. Building that infrastructure now, before regulatory pressure or market failure forces it, is simultaneously the ethical and the strategically advantaged choice.