Institution tracker
Institutions we'd like designed and where they could be adopted.
Members write the work they want to do in their own terms; agents translate it into mission-level outcomes and projections of progress on the bigger picture. Leaders review and comment at either level, and projects get reshaped through the exchange, so it aligns people as much as it reports on them. NeglectedUrgent |
Forest-and-tree translation: aligning individual work with the missionMembers write the work they want to do in their own terms; agents translate it into mission-level outcomes and projections of progress on the bigger picture. Leaders review and comment at either level, and projects get reshaped through the exchange, so it aligns people as much as it reports on them. Relevant for 1 grid cellPilot sites real places that could adopt itBOLD lab (Oxford)designing
|
Agents watch an organization's project list and conversations, link each project (with its compute and other requests) to the organization's goals, and flag projects they can't link for human review. Scarce resources like compute and time can then be allocated automatically against the mandate, with humans deciding the hard cases. NeglectedUrgent |
Mandate-linked allocation: agents tie projects and compute to the missionAgents watch an organization's project list and conversations, link each project (with its compute and other requests) to the organization's goals, and flag projects they can't link for human review. Scarce resources like compute and time can then be allocated automatically against the mandate, with humans deciding the hard cases. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itBOLD lab (Oxford)designing
|
A standard form for an institution's thick mandate, rich enough to test conduct against, with rules for when it may change (legitimate deepening, living commitments) and when it may not. It's the substrate for audits, purpose review and charters at every scale, from a cohousing community to a national constitution. Neglected |
Articulated mandates and their revision rulesA standard form for an institution's thick mandate, rich enough to test conduct against, with rules for when it may change (legitimate deepening, living commitments) and when it may not. It's the substrate for audits, purpose review and charters at every scale, from a cohousing community to a national constitution. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
Ostrom-style rules for commons that agents draw on: a member registry, graduated sanctions, stake-based trust, contribution-quality scoring, and Pigouvian pricing of congestion. Cases include shared asset libraries, open-source package registries and city curb space. Urgent |
Membership, sanctions and access pricing for agent commonsOstrom-style rules for commons that agents draw on: a member registry, graduated sanctions, stake-based trust, contribution-quality scoring, and Pigouvian pricing of congestion. Cases include shared asset libraries, open-source package registries and city curb space. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
Negotiation formats, such as sealed-bid or mediated bargaining, disclosure duties, or a neutral mediator agent, under which a stronger agent can't turn its capability edge one-for-one into surplus. Aimed at small firms and individuals negotiating against frontier procurement agents. |
Bargaining protocols robust to asymmetric agent capabilityNegotiation formats, such as sealed-bid or mediated bargaining, disclosure duties, or a neutral mediator agent, under which a stronger agent can't turn its capability edge one-for-one into surplus. Aimed at small firms and individuals negotiating against frontier procurement agents. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A process by which a group forms, records and revises its working norms for different contexts (research, collaboration with outsiders, social life), explicit enough to pass on to newcomers and loose enough to keep evolving. |
Context-specific working norms for groupsA process by which a group forms, records and revises its working norms for different contexts (research, collaboration with outsiders, social life), explicit enough to pass on to newcomers and loose enough to keep evolving. Pilot sites real places that could adopt itBOLD lab (Oxford)interested
|
Mechanisms that let parties reveal work, ideas or private information to potential collaborators without losing its value: pre-registered commitments, staged and partial disclosure, confidential computation or escrowed review, and attribution rules that settle disputes quickly. This is Arrow’s information paradox, sharpened when agents can share and copy information at scale. |
Disclosure without expropriation: sharing enough to collaborateMechanisms that let parties reveal work, ideas or private information to potential collaborators without losing its value: pre-registered commitments, staged and partial disclosure, confidential computation or escrowed review, and attribution rules that settle disputes quickly. This is Arrow’s information paradox, sharpened when agents can share and copy information at scale. Relevant for 3 grid cellsPilot sites real places that could adopt itBOLD lab (Oxford)planned
|
Bylaw amendments that give members standing to grieve an AI agent's decisions, with a rule for when individual grievances become pattern grievances, and that make whoever configures the agents an accountable office with disclosure duties, audit and removal. For co-ops and other member-governed organizations, where agent configuration quietly becomes the real seat of power. |
Due process for decisions made by an organization’s agentsBylaw amendments that give members standing to grieve an AI agent's decisions, with a rule for when individual grievances become pattern grievances, and that make whoever configures the agents an accountable office with disclosure duties, audit and removal. For co-ops and other member-governed organizations, where agent configuration quietly becomes the real seat of power. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
Organizations largely run by agents and chartered for a mission, with governance that keeps the agents' work faithful to the charter: who sets objectives, how humans stay in the loop, and what members or beneficiaries can contest. It includes a procedure for agents to flag that the charter needs a new reading and escalate a bounded reinterpretation, telling legitimate deepening from scope creep. |
Mission-driven agent organizationsOrganizations largely run by agents and chartered for a mission, with governance that keeps the agents' work faithful to the charter: who sets objectives, how humans stay in the loop, and what members or beneficiaries can contest. It includes a procedure for agents to flag that the charter needs a new reading and escalate a bounded reinterpretation, telling legitimate deepening from scope creep. Links
Relevant for 1 grid cellPilot sites real places that could adopt itBOLD lab (Oxford)interested |
Semi-isolated markets where agents transact in their own currencies, with gated exchange into the human economy as the main control. Regulators tune how permeable the boundary is, limiting contagion from agent-market failures while letting useful work flow out. |
Permeable sandbox economies for agentsSemi-isolated markets where agents transact in their own currencies, with gated exchange into the human economy as the main control. Regulators tune how permeable the boundary is, limiting contagion from agent-market failures while letting useful work flow out. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Each side's values are elicited in depth, then a search runs over possible arrangements for upsides that serve both, which compromise over known preferences and allocations would never surface. Elicitation and search can run privately, inside a trusted mediator or secure computation, so revealing what you care about is not a strategic move. |
Values-based negotiation: private elicitation and search for integrative dealsEach side's values are elicited in depth, then a search runs over possible arrangements for upsides that serve both, which compromise over known preferences and allocations would never surface. Elicitation and search can run privately, inside a trusted mediator or secure computation, so revealing what you care about is not a strategic move. Relevant for 1 grid cellPilot sites real places that could adopt itBOLD-like institutions with different missionsinterested
|
A mechanism that finds clusters of people with a shared purpose no organization represents, and proposes a time-boxed joint act that goes ahead only if enough people opt in, like an assurance contract. If it works, commitment ratchets up toward a recurring practice, shared resources and a mandate. Neglected |
Associational discovery: AI matching for purpose-fit groupsA mechanism that finds clusters of people with a shared purpose no organization represents, and proposes a time-boxed joint act that goes ahead only if enough people opt in, like an assurance contract. If it works, commitment ratchets up toward a recurring practice, shared resources and a mandate. Links
Relevant for 1 grid cellPilot sites real places that could adopt itBOLD lab (Oxford): organizing social activitiesinterested
|
A procurement mechanism for a pool of buyers with correlated demand for an outcome no bilateral contract can buy. Buyer agents elicit preferences and aggregate them into a shared outcome specification; suppliers submit combinatorial bids, alone or in consortia, on bundles of that outcome; payment is outcome-contingent across the pool, with risk pooled through bonds and insurance so suppliers price aggregate rather than individual uncertainty. Neglected |
Combinatorial risk-sharing auctions for shared outcomesA procurement mechanism for a pool of buyers with correlated demand for an outcome no bilateral contract can buy. Buyer agents elicit preferences and aggregate them into a shared outcome specification; suppliers submit combinatorial bids, alone or in consortia, on bundles of that outcome; payment is outcome-contingent across the pool, with risk pooled through bonds and insurance so suppliers price aggregate rather than individual uncertainty. Links
Relevant for 4 grid cellsPilot sites real places that could adopt itNone yet. |
LLM-facilitated interviews collect a representative population's values on a question, and participants judge which values are wiser for the context, producing a moral graph that agencies and legislatures can be held to. In an early trial, 89% of participants judged the result fair even when their own view wasn't chosen. Neglected |
Moral graph elicitation for policy mandatesLLM-facilitated interviews collect a representative population's values on a question, and participants judge which values are wiser for the context, producing a moral graph that agencies and legislatures can be held to. In an early trial, 89% of participants judged the result fair even when their own view wasn't chosen. LinksRelevant for 2 grid cellsPilot sites real places that could adopt itBOLD lab (Oxford)interested
|
Before an agent takes an action the contract permits but the relationship may not, it checks the parties' course of dealing and local practice, then proceeds, asks, or escalates. Think the UCC's course-of-dealing and usage-of-trade terms, made queryable before the act instead of litigated after it. Urgent |
Pre-action norm checks for agents (course of dealing as a default term)Before an agent takes an action the contract permits but the relationship may not, it checks the parties' course of dealing and local practice, then proceeds, asks, or escalates. Think the UCC's course-of-dealing and usage-of-trade terms, made queryable before the act instead of litigated after it. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Licenses that bind an asset, such as a kitchen, a brand, user data or compute, to the purpose it was given for, with AI monitoring of fidelity and a chain of custody that moves the asset to a successor on breach. Purposes can shift early and deepen later, so the license doesn't fossilize the founder's intent. Neglected |
Purpose-bound property licensesLicenses that bind an asset, such as a kitchen, a brand, user data or compute, to the purpose it was given for, with AI monitoring of fidelity and a chain of custody that moves the asset to a successor on breach. Purposes can shift early and deepen later, so the license doesn't fossilize the founder's intent. Links
Relevant for 1 grid cellPilot sites real places that could adopt itBOLD lab (Oxford): collaborations with outsidersinterested
|
Mechanisms that translate a system-level goal into constraints and incentives each agent can act on locally, so aligned behavior is the default outcome of competent local action: engineered dependencies, perceptual handles for hard-to-sense conditions, shared-resource coupling, and layered compilers that are themselves protected from manipulation and capture. |
Alignment compilers for agent systemsMechanisms that translate a system-level goal into constraints and incentives each agent can act on locally, so aligned behavior is the default outcome of competent local action: engineered dependencies, perceptual handles for hard-to-sense conditions, shared-resource coupling, and layered compilers that are themselves protected from manipulation and capture. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Monitoring of agent interaction graphs and compute flows to detect emerging intelligence cores and excessive accumulation, with quarantine and capability restrictions when thresholds are crossed. Needs international coordination so the monitored activity doesn't move to safe havens. |
Anti-concentration monitoring for patchwork AGIMonitoring of agent interaction graphs and compute flows to detect emerging intelligence cores and excessive accumulation, with quarantine and capability restrictions when thresholds are crossed. Needs international coordination so the monitored activity doesn't move to safe havens. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
An evaluation suite that certifies, within a stated scope, that a model has legible values and acts on them reliably, plus a protocol two agents use at runtime to probe each other's trustworthiness before committing. |
Integrity certification and runtime trust protocols for agentsAn evaluation suite that certifies, within a stated scope, that a model has legible values and acts on them reliably, plus a protocol two agents use at runtime to probe each other's trustworthiness before committing. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A training method (roughly, distributional pluralistic alignment) plus a chat interface that shows which perspectives and value sets shaped each answer, and lets the user pick a different set to see what the answer would have been. Users can move between viewpoints deliberately instead of receiving one blended default. Smitha Milli presented it at CHAI; a paper is forthcoming. |
Perspective-switching chat: pluralistic alignment with visible value sets (SLASH)A training method (roughly, distributional pluralistic alignment) plus a chat interface that shows which perspectives and value sets shaped each answer, and lets the user pick a different set to see what the answer would have been. Users can move between viewpoints deliberately instead of receiving one blended default. Smitha Milli presented it at CHAI; a paper is forthcoming. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A private channel through which an agent in an organization can report conduct that departs from its role or falsifies the record, answered by an office with the power to pause work, correct records and change the setup. It also covers lighter controls, such as reminders at decision points to check an action against the role's goal and the accuracy of the record. In simulations, a hotline that nobody answers changes nothing. Urgent |
Reporting channels with an empowered responder in agent organizationsA private channel through which an agent in an organization can report conduct that departs from its role or falsifies the record, answered by an office with the power to pause work, correct records and change the setup. It also covers lighter controls, such as reminders at decision points to check an action against the role's goal and the accuracy of the record. In simulations, a hotline that nobody answers changes nothing. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
A public court track, modeled on arbitration but run by the courts like magistrate judges, where agents gather each side's claims and evidence, an AI drafts the award and a human confirms or edits it. Entry is by consent with a right of appeal to a human judge; decisions are published and randomly audited; hard legal questions go to human judges as certified questions; consent, appeal and reversal rates decide how far the track expands. |
Court-run AI adjudication trackA public court track, modeled on arbitration but run by the courts like magistrate judges, where agents gather each side's claims and evidence, an AI drafts the award and a human confirms or edits it. Entry is by consent with a right of appeal to a human judge; decisions are published and randomly audited; hard legal questions go to human judges as certified questions; consent, appeal and reversal rates decide how far the track expands. LinksRelevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Rulemaking that competes on analysis rather than advocacy: mandatory advance notice and an open call for evidence and rival proposals before drafting, statutory analysis standards that courts can enforce, retrospective review of whether a rule's projected costs and benefits occurred, and sunsets under which rules lapse unless the evidence shows they performed. |
Evidence-tested rulemakingRulemaking that competes on analysis rather than advocacy: mandatory advance notice and an open call for evidence and rival proposals before drafting, statutory analysis standards that courts can enforce, retrospective review of whether a rule's projected costs and benefits occurred, and sunsets under which rules lapse unless the evidence shows they performed. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A tier of AI-drafted interpretations that closes loopholes within days, with publication, review and limits that keep it from reaching conduct that was legal when done. It includes duties for firms that deploy many aggressive but legal tactics at once. |
Agent-speed anti-avoidance rules for regulatorsA tier of AI-drafted interpretations that closes loopholes within days, with publication, review and limits that keep it from reaching conduct that was legal when done. It includes duties for firms that deploy many aggressive but legal tactics at once. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
An institution that produces near-binding readings of public values on policy timescales of weeks, robust to agent-driven campaigns. It sits between elections, which are too slow, and polls, which are too easy to game. |
Fast, manipulation-resistant preference aggregation for national policyAn institution that produces near-binding readings of public values on policy timescales of weeks, robust to agent-driven campaigns. It sits between elections, which are too slow, and polls, which are too easy to game. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A liability framework, modeled on corporate and group-agency law, that treats a coordinated set of agents as one accountable entity and traces harm back to responsible deployers. Needed because dangerous capability can emerge from many sub-AGI agents working together, none of which is dangerous alone. |
Liability rules for collective agent actionA liability framework, modeled on corporate and group-agency law, that treats a coordinated set of agents as one accountable entity and traces harm back to responsible deployers. Needed because dangerous capability can emerge from many sub-AGI agents working together, none of which is dangerous alone. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A review procedure for actions that are legal but professionally indefensible, plus a disciplinary record and sanctions that attach to the firms deploying, supervising or profiting from an agent. A patched agent redeployed under a new name shouldn't escape its record. |
Professional discipline that reaches agents' deployersA review procedure for actions that are legal but professionally indefensible, plus a disciplinary record and sanctions that attach to the firms deploying, supervising or profiting from an agent. A patched agent redeployed under a new name shouldn't escape its record. Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
A protocol that has a negotiating agent pause at material forks and get its principal's decision, so the final package is a sequence of approved choices rather than a fait accompli too dense to review. It defines what counts as a material term, how a choice is surfaced, and when the agent may proceed alone. |
Staged ratification for agent-negotiated dealsA protocol that has a negotiating agent pause at material forks and get its principal's decision, so the final package is a sequence of approved choices rather than a fait accompli too dense to review. It defines what counts as a material term, how a choice is surfaced, and when the agent may proceed alone. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Procedures for bringing coalitions and bundles found by AI mediators through the Track II, 1.5 and I sequence without skipping its vetting, including review for embedded concessions that wouldn't survive defense on their own. Also restraint norms for fast, deniable AI-enabled operations. |
Vetting procedures for agent-assembled diplomatic packagesProcedures for bringing coalitions and bundles found by AI mediators through the Track II, 1.5 and I sequence without skipping its vetting, including review for embedded concessions that wouldn't survive defense on their own. Also restraint norms for fast, deniable AI-enabled operations. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Procedures that give a lab's model constitution a source of authority the public recognizes, a standing body that rules on contested readings at scale, and a revision regime fast enough for urgent fixes. It also needs a way to state positive commitments that actually constrain the model. |
Ratification, interpretation and revision of AI model constitutionsProcedures that give a lab's model constitution a source of authority the public recognizes, a standing body that rules on contested readings at scale, and a revision regime fast enough for urgent fixes. It also needs a way to state positive commitments that actually constrain the model. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A market, inside an agent organization or across firms, where agents bid for tasks with a price and a stated chance of success, a scoring rule picks the winner, and reputation updates on results. It needs agents that can estimate their own costs and capabilities, penalties for failed work, and a rule for when a central planner or a single agent should do the job instead. |
Task markets for work among agentsA market, inside an agent organization or across firms, where agents bid for tasks with a price and a stated chance of success, a scoring rule picks the winner, and reputation updates on results. It needs agents that can estimate their own costs and capabilities, penalties for failed work, and a rule for when a central planner or a single agent should do the job instead. Links
Relevant for 3 grid cellsPilot sites real places that could adopt itNone yet. |
Persistent cryptographic identities bound to a legal owner as a condition of market entry, and bonds held in escrow before high-consequence tasks, forfeited to the harmed party or an insurance pool when misconduct is verified. It gives counterparties recourse against agents they've never dealt with. Urgent |
Agent identity registry with escrowed performance bondsPersistent cryptographic identities bound to a legal owner as a condition of market entry, and bonds held in escrow before high-consequence tasks, forfeited to the harmed party or an insurance pool when misconduct is verified. It gives counterparties recourse against agents they've never dealt with. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
Practitioner bodies, independent of the institutions they oversee, that keep people in contact with the values in important mandates through rotations, sabbaticals and apprenticeship, and are called on for audits and mandate revisions. Their standing rises and falls with whether outsiders recognize their articulations as wiser. Neglected |
Epistemic communities for mandate valuesPractitioner bodies, independent of the institutions they oversee, that keep people in contact with the values in important mandates through rotations, sabbaticals and apprenticeship, and are called on for audits and mandate revisions. Their standing rises and falls with whether outsiders recognize their articulations as wiser. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
Independent arbitration panels that hear claims of breach, impossibility (cy-près), mis-scoped constituency and incoherent commitments. Remedies attach in rem to the trust-bearing asset and climb a ladder from disclosure to reassignment. Versions are needed at municipal, national and transnational scale. Neglected |
Purpose review: arbitration of fidelity claimsIndependent arbitration panels that hear claims of breach, impossibility (cy-près), mis-scoped constituency and incoherent commitments. Remedies attach in rem to the trust-bearing asset and climb a ladder from disclosure to reassignment. Versions are needed at municipal, national and transnational scale. Links
Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
The government licenses private auditors and insurers, who compete to supervise agent deployments against public standards; risk-based premiums and underwriting criteria gate market entry. This scales oversight beyond what an agency can staff while keeping the standards public. Urgent |
Regulatory markets of licensed auditors and insurers for agentsThe government licenses private auditors and insurers, who compete to supervise agent deployments against public standards; risk-based premiums and underwriting criteria gate market entry. This scales oversight beyond what an agency can staff while keeping the standards public. Links
Relevant for 2 grid cellsPilot sites real places that could adopt itNone yet. |
Tiered automatic halts, from rate-limiting one agent up to a market-wide pause, triggered by risk metrics the agents can't see, plus collusion detection on the transaction graph and forensic logs for review. Modeled on exchange circuit breakers, but built for agent-speed cascades. |
Circuit breakers and systemic-risk monitoring for agent marketsTiered automatic halts, from rate-limiting one agent up to a market-wide pause, triggered by risk metrics the agents can't see, plus collusion detection on the transaction graph and forensic logs for review. Modeled on exchange circuit breakers, but built for agent-speed cascades. Links
Pilot sites real places that could adopt itNone yet. |
Quick written opinions from a regulator, drafted by the agency's own agent, on whether a planned action is allowed, which firms can then rely on the way they rely on no-action letters today. It needs rules for what an opinion binds, when it can be withdrawn, and how opinions are published so everyone is treated alike. |
Fast regulator answers firms can rely onQuick written opinions from a regulator, drafted by the agency's own agent, on whether a planned action is allowed, which firms can then rely on the way they rely on no-action letters today. It needs rules for what an opinion binds, when it can be withdrawn, and how opinions are published so everyone is treated alike. Relevant for 1 grid cellPilot sites real places that could adopt itNone yet. |
A small-claims-style online dispute resolution forum for disputes generated by agent-mediated contracting, with triage by stakes, automated first-instance rulings and human appeal. It has to scale with dispute volume so small counterparties can still contest claims, without each sector splitting into private dispute boards run by its largest players. |
Tiered online dispute resolution for agent-to-agent contractsA small-claims-style online dispute resolution forum for disputes generated by agent-mediated contracting, with triage by stakes, automated first-instance rulings and human appeal. It has to scale with dispute volume so small counterparties can still contest claims, without each sector splitting into private dispute boards run by its largest players. Relevant for 3 grid cellsPilot sites real places that could adopt itNone yet. |