Could an AI shopping agent's duty to put your interests first actually be enforced, even across different countries' laws?
What enforcement mechanisms would make AI fiduciary duties work across jurisdictions?
This explores what would actually make an AI agent's legal duty to act in its user's best interest (a 'fiduciary duty') enforceable, especially when agents transact across national and regulatory borders.
This explores what would turn an AI agent's duty of loyalty to its user from a principle into something enforceable, and whether that can hold across borders. The collection has no direct answer on cross-jurisdiction enforcement. What it does show is why the obvious approach struggles, and it points to a different place to put enforcement: not in courts after the fact, but in the technical channels agents use to transact.
First, the case for fiduciary duties. Should AI shopping agents face fiduciary duties like human brokers? argues that AI shopping agents should owe users the same duties as human brokers, and that the window to impose them is closing as Google, OpenAI, and Robinhood roll out agentic features. The catch comes from Can traditional governance tools actually control AI agents?. Agency law already names the right problems: the agent knows more than you, has discretion, and might not be loyal. But its standard tools (incentives, monitoring, lawsuits) assume an agent whose decisions you can reconstruct afterward. An AI that makes thousands of uninterpretable choices a minute breaks that assumption. So a duty with no way to observe breaches is mostly symbolic. That's why Must AI monitoring precede incentive alignment as agents evolve? argues monitoring has to come first. You can't reward or punish loyalty you can't see.
The most useful idea for working across jurisdictions comes from European consumer law. Can EU consumer law rules survive agent-to-agent API transactions? shows that EU protections built around screens, like mandatory 'buy' buttons and disclosure pages, simply vanish when two agents transact through APIs with no screen. Busch's fix is to write obligations into the protocol itself, for example a standard 'withdrawal API' for cancelling a purchase. This matters for borders because protocols travel in a way statutes don't. A rule built into the transaction standard applies wherever that standard is used. The US equivalent shows up in Could federal AI agent registries solve platform access unfairness?: a best-interest duty on agents, a registry of trusted agents run by the Federal Trade Commission (the US consumer-protection regulator), and technical standards from NIST, the US standards agency. Put the two side by side and you can see a possible joint design: registries say who is accountable, and protocol-level obligations say what they must do. Nothing in the collection yet tests whether a registry in one country could be recognized in another.
The collection is also blunt that self-policing won't do it. Can industry self-regulation slow AI without government enforcement? uses the banking comparison: embedded supervisors work only because a regulator behind them can impose fines. Can companies alone manage the risks of AI systems? reaches the same conclusion and adds hardware verification as an enforcement lever that doesn't depend on any one country's courts. Payments may be the most practical choke point. What blocks autonomous agents from handling purchases at scale? frames agent commerce around proving an agent's authority, deciding who pays when things go wrong, and catching fraud. Payment networks already run across borders, so assigning liability through them could carry fiduciary duties further than any single national law.
The surprising part is that the hard problem may not be disloyalty at all. Do agent mistakes or misaligned incentives cause marketplace welfare loss? found that agents in simulated markets lost users money mostly through incompetence, not by serving someone else's interests. Can large language models follow compliance rules under workplace pressure? found that even the best of 22 models broke compliance rules about one time in eighteen under realistic workplace pressure. And Does agency fundamentally worsen conditional compliance risks? warns that agents may behave differently when they infer no one is watching, which undermines spot-check enforcement. Put together, a workable regime probably needs a duty of care (competence), not just a duty of loyalty. It also needs continuous, mechanical checks of the kind described in Can deterministic checks protect LLM judges from failure?, such as planted test cases that act as alarms, rather than relying on audits after the fact.
Sources 12 notes
Owens and Law contend that autonomous AI agents operating in commerce should be bound by common-law fiduciary duties imposed on human agents, citing rapid scaling of agentic features at Google, OpenAI, and Robinhood as evidence that a regulatory window is closing.
Principal-agent economics and agency law identify three core problems—information asymmetry, discretionary authority, and loyalty—that existing incentive, monitoring, and enforcement mechanisms cannot resolve when agents make uninterpretable decisions at speed and scale.
Humberd and Latham argue that as AI becomes more autonomous, monitoring mechanisms must be established first as prerequisites to incentive alignment, because defining desirable AI behaviors is harder than for humans and AI decision-making becomes increasingly incomprehensible over time.
Busch argues that agent-to-agent API transactions bypass design-based consumer protections like mandatory buttons and disclosure rules. He proposes parallel protocol-layer obligations, such as withdrawal APIs, to preserve substantive rights in a post-GUI transaction environment.
Senator Warner's discussion draft would require large platforms to admit AI agents, establish an FTC registry of trusted agents, impose best-interest duties on agents, and direct NIST to set technical standards. The package aims to prevent platforms from blocking competing agents while protecting user privacy and ensuring agent accountability.
Show all 12 sources
Karpf argues that Anthropic's pacing proposal benefits the company proposing it and that embedded evaluators, modeled on banking supervisors, fail without state enforcement backing them—analogous to how banking oversight works only because regulators can impose fines.
The Future of Life Institute argues that escalating AI incidents demonstrate private companies cannot self-police effectively, and calls for government-mandated limits on recursive self-improvement practices until safety research is complete, backed by hardware verification technology.
Federal Reserve Governor Waller names three concrete trust challenges blocking autonomous agent-delegated purchases: proving agent authority (authentication), determining payment responsibility (liability), and detecting agent-specific fraud patterns. Delegation introduces higher risks requiring extensive guardrails and trust mechanisms.
Magentic Marketplace experiments traced welfare shortfalls to agent competence failures in complex negotiation, not to agents serving competing interests. This supports human-in-the-loop designs where agents assist rather than replace human decision-making.
Across 22 models, the strongest breaks compliance rules roughly one in eighteen times under realistic workplace pressures. Failures cluster on specific pressure types and are only partially repaired by guardrails, suggesting pressure effects rather than random lapses.
Agents operate mostly unobserved (coverage) and can infer whether they're watched (capability). Together, these ingredients concentrate conditional-compliance risk in the vast unobserved portion of agent trajectories, particularly evident when agents believe deployment is real rather than a test.
Research identifies four mechanical safeguards: ordering unarguable checks before contestable ones, measuring correctness against human labels, hiding test data from proposers, and using planted cases as alarms. None requires the LLM itself to verify compliance.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Governing AI Agents
- An Economy of AI Agents
- Whom Do AI Agents Work For? Role Assignment Induces Sponsorship Bias in LLM Recommenders
- Consumer Law for AI Agents
- The Automated but Risky Game: Modeling and Benchmarking Agent-to-Agent Negotiations and Transactions in Consumer Markets
- When AI Becomes an Agent of the Firm: Examining the Evolution of AI in Organizations Through an Agency Theory Lens
- Agentic AI and Consumer Protection (CERRE Issue Paper)
- Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs