The Global Risks Forum (GRF)
GRF · AI and Digital Governance
Situational awarenessEvents, outlooks and readiness for those who decide
AI agents now plan and take actions across software, payments and physical systems with limited human input. When an agent causes harm, existing liability rules must decide who pays: the developer, the deployer, the user or the operator of the platform it acted on. This page sets out how autonomy strains liability and oversight, the records that show it, and who decides how responsibility is assigned.
Action boundaries and accountability. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
An agent pursuing a goal can take steps its user did not foresee or authorise. Fault-based liability struggles when no person intended or could reasonably foresee the harmful action.
The model, the agent framework, tool providers, the deploying firm and the user all contribute. Proving which defect caused harm is costly for claimants without access to records.
What an agent can access, spend or change sets the size of possible harm. Limits, confirmations and kill switches are governance choices made at deployment.
Product liability, contract and tort law are being reinterpreted for software and AI. Until courts and legislatures settle questions, uncertainty itself shapes deployment.
The working screen behind this subject: its structure, terms and states. Names are placeholders and values illustrative.
Autonomous agents and liability · Algorithm register · Quarter 2
| Benefit fraud flagging | Social agency | High | Missing | Partial |
| Permit triage | Planning | Limited | Done | Yes |
| School placement | Education | High | Done | Yes |
| Chat assistant | Tax office | Minimal | Done | n/a |
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| Jurisdictions treating software as a product for liability | count | on change | Legislation | Published statistics |
| Court decisions on AI-caused harm | count by type | annual | Case law databases | Published statistics |
| Agent incidents reported | count by harm type | monthly | Incident databases | REPORTED |
| Deployments with documented permission limits | share of surveyed deployments | periodic | Surveys and audits | DECLARED |
| Insurance products covering AI liability | count of offerings | annual | Market reports | REPORTED |
| Regulatory guidance on agentic AI | count of documents | annual | Regulator publications | Published statistics |
The instruments, mandates, actors and accountability arrangements a policymaker, diplomat or development agency works with, citing the agreements that set them. Structures follow the frameworks named. Any state or progress drawn is illustrative, not a record.
The kinds of instrument available on autonomous agents and liability, by level of government and cooperation.
| Instrument type | Local | National | Regional | Global |
|---|---|---|---|---|
| Regulatory | Zoning, building codes | Laws, standards, licensing | Harmonised rules | Conventions and treaties |
| Economic | Local levies, grants | Budget lines, taxes, insurance schemes | Risk pools, regional funds | Financing mechanisms |
| Information | Local warnings | Disclosure, early warning, statistics | Data sharing arrangements | Global monitoring and reporting |
| Institutional | Local committees | Lead agencies, coordination councils | Regional bodies and secretariats | Multilateral mandates |
| Cooperative | Community agreements | Partnership compacts | Memoranda and joint programmes | Agreements and joint commitments |
The topic as a computational graph: named sources feed the methods, an evidence review gates the outputs, and the decisions stay with the authorities that hold them.
Fig. 02Nodes carry their type, two metric rows and a state; connectors show what feeds what; dashed frames name the owners; the timeline is the cadence. Design of the pipeline, not a record of a run.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| EU revised Product Liability Directive | Liability rules including software and AI systems as products | Public, EU reuse terms |
| Law commission and parliamentary reports on AI liability | Analysis of liability options for AI and automated systems | Public records, terms per body |
| OECD AI Incidents Monitor | Recorded incidents involving AI systems | Free, OECD terms |
| UK Automated and Electric Vehicles Act and Automated Vehicles Act | Liability allocation for automated vehicles | Open Government Licence |
| Court judgment databases | Decisions on harm involving automated systems | Public records, terms per court |
AI and Digital Governance prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Legislatures and law commissions | How liability rules apply to autonomous systems | Comparative regimes and case evidence |
| Deploying organizations | What an agent may do and what controls apply | Control practice and incident evidence |
| Insurers | Whether and how to cover AI liability | Incident patterns and legal developments |
| Courts | Who is liable for harm in a specific case | Records presented by parties; the page serves context only |
Most governance frameworks were built for systems that advise a person who then acts. Agents collapse that gap, and with it one of the main points at which errors were caught. Why now: AI agents that act on behalf of people and organisations raise questions of liability that existing rules did not foresee, and legislators are still working through them. The platform is designed to compare emerging approaches and to record open questions for decision holders, without giving legal advice.
Agents can act faster and more often than people can review, so errors can multiply before anyone notices.
Interactions between automated systems can produce outcomes no one designed, as seen in automated trading.
Existing liability rules assume a human or company acted; agents complicate who that is.
Technical limits on what agents can access and do, studied under AI evaluation and assurance engineering. Limits: agents may find unintended routes around controls.
Recording actions and designing them to be undoable where possible. Limits: some actions cannot be reversed.
Requiring human sign-off above defined risk levels. Limits: can slow operations and invite approval fatigue.
Comparing how legal systems allocate responsibility. Limits: law is still developing.
Nexus Reports is the Consortium's quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
Never send personal data, sensitive infrastructure detail or anything you have no right to share. A contribution is reviewed before use; sending it confers no authorship or editorial rights, and sponsorship never buys a finding.
The current service level is R0: information and enquiries. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.
Maps, data and figures on this site are shown for context and general information. They are not advice, forecasts or warnings; the limits of the information are set out in the Disclaimers and Data Notices.