GCRI · Inteligencia artificial y sistemas autónomos
Sistemas agénticos
Agentic AI sistemas pursue goals by planning and taking actions: calling tools, browsing, writing and running code, or operating other software, sometimes over long tasks with little supervision. Autonomy brings usefulness and new failure modes, from compounding errors to misuse of granted permissions. This page sets out how agents fail, how their actions can be bounded and overseen, and what evidence serves those who build, deploy and govern them.
Agentic systems on the record.
Agentic systems: how the risk develops.
Tool access, delegated authority and safe execution. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Errors compound over steps
Each step in a long task can go wrong, and an early mistake shapes later actions. Small per-step error rates add up over many steps.
Permissions define the blast radius
An agent can only do harm with the access it has been given. Broad credentials, write access and payment rights turn errors or manipulation into real-world consequences.
Untrusted inputs steer behaviour
Agents read web pages, documents and messages that may contain instructions. Prompt injection can redirect an agent to act against its user.
Oversight must be designed in
Humans cannot review every action of a fast agent. Approval steps for consequential actions, logging and the ability to stop and roll back are design choices.
Agent oversight console.
La pantalla que utilizaría un equipo para seguir esta materia: sus mediciones, estados y decisiones. Los nombres son ejemplos y los valores son ilustrativos.
Agent oversight console
Agentic systems · Agent sandbox · Session AG-6627
Tool calls
- 10:34Infosearch(documents) · read-only
- 10:51Inforun_model(flood_nowcast) · sandbox
- 10:51Hitosend_message(external) · needs approval
- 10:53Errorwrite_file(outside workspace) · denied
Política
- Alcance
- workspace only
- External actions
- human approval
- Spend limit
- 43 units
- Kill switch
- Armed
Actions by approval class
Agents that act need guardrails in every domain they touch.
- EnergíaAcoplamientoAutomated control of grids needs strict human oversight.What to watchOverride and rollback design
- SaludAcoplamientoAgents handling health data raise privacy and safety risks.What to watchAccess scopes and logs
- AguaAcoplamientoAutomation of utility operations must fail safe.What to watchFail-safe behaviour
Agentic system studies are designed for the serie Exponential Technology. The OWASP Top 10 for LLM applications lists failure modes such as excessive agency and prompt injection; the page describes them without exploit detail.
- Incident reports involving agents.
- Designs for oversight and permission scoping.
How agentic systems is modelled, measured and tested.
El modelo, sus supuestos, sus datos y su incertidumbre, en la forma en que un científico o un ingeniero los comprobaría. Las estructuras siguen métodos publicados. Los valores representados son ilustrativos, no hallazgos.
Modelo de sistema
The couplings between the systems in agentic systems: signs of influence, a reinforcing and a balancing loop, and cross-links where stress jumps.
Qué se mide y con qué calidad.
Cada indicador lleva su unidad, su periodicidad, el tipo de fuente que lo respalda y la etiqueta de evidencia que llevaría un valor. No se muestran valores: un programa los aporta bajo su propia autoridad.
| Indicador | Unidad | Cadencia | Tipo de fuente | Etiqueta |
|---|---|---|---|---|
| Task success rate on agent benchmarks | percent of tasks completed correctly | per evaluation | Agent benchmark results | NOTIFICADO |
| Actions requiring human approval | percent of consequential actions | per deployment | Deployment configuration | DECLARADO |
| Prompt injection success rate | percent of attempts succeeding | per evaluation | Security testing | OBSERVADO |
| Agent incidents | count by severity | mensual | Incident reports | NOTIFICADO |
| Actions logged and attributable | percent of agent actions | per deployment | Audit logs | OBSERVADO |
| Mean task length before human intervention | steps or minutes | per evaluation | Evaluation logs | OBSERVADO |
De dónde procede la evidencia.
Conjuntos de datos públicos y registros institucionales identificados, cada uno con su tipo de licencia. Las fuentes obtenidas por acuerdo solo se utilizan conforme a sus condiciones y nunca se republican.
| Fuente | Lo que aporta | Tipo de licencia |
|---|---|---|
| OWASP Top 10 for LLM Applications and agentic security guidance | Risks such as prompt injection and excessive agency | Open, CC BY-SA 4.0 |
| Published agent benchmarks such as SWE-bench and related suites | Task-based evaluation of agents | Open, licence per benchmark (to check) |
| NIST AI Risk Management Framework | Practices for governing and measuring AI risk | Public domain, US government |
| AI Incident Database | Reported incidents involving AI sistemas | Open, licence to check |
| Developer system cards describing agent safeguards | Developer-reported agent evaluations and controls | Public documents, terms per developer |
Quién decide y a qué sirve la evidencia.
Estudio de Inteligencia artificial y sistemas prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Quién decide | La pregunta | Evidencia a la que sirve esta página |
|---|---|---|
| Deploying organisations | Which actions an agent may take alone and which need approval | Permission patterns; evaluation results |
| Security teams | How to protect agents from manipulation | Injection testing; incident evidence |
| Reguladores | What oversight to require for autonomous AI actions | Oversight design evidence |
Productos que genera este tema
- Permission and approval design patterns for agents
- Agent evaluation protocols covering long tasks
- Prompt injection test suites for deployments
- Logging and rollback requirements
What designers and evaluators ask.
- What actions can the agent take, with whose authority, and which are reversible?
- How does it behave when instructions are ambiguous, conflicting or malicious?
- Can a human see what it is doing, understand why and stop it in time?
- How do errors compound over long sequences of steps?
How agents fail.
An agent combines a model’s reasoning with permissions in the world. Small errors in early steps can compound across a long task. Agents can misinterpret goals, pursue them in unintended ways or take shortcuts that satisfy a metric but not the intent. Content the agent reads, such as a web page or document, can contain instructions designed to hijack it, an attack known as prompt injection.
The more permissions an agent holds, the larger the potential harm. sistemas that coordinate several agents add further failure modes, as agents pass errors to each other or interact in unforeseen ways.
Designing for safe execution.
- 01
Least privilege
Granting only the permissions a task needs, with sandboxes and spending or action limits, drawing on ciberseguridad y criptografía.
- 02
Human oversight points
Requiring confirmation for consequential or irreversible actions. Too many prompts lead to rubber-stamping, so placement matters.
- 03
Task-based evaluation
Testing agents on realistic multi-step tasks in controlled environments, using Ingeniería de evaluación y garantía de la IA. Simulated environments miss some real-world complexity.
- 04
Logging and audit
Recording actions and reasoning traces so incidents can be reconstructed. Traces may not fully reflect how a model reached a decision.
Cómo se estudia y lo que los métodos no pueden decir.
- 01Ingeniería de evaluación y garantía de la IACasos de aseguramiento para sistemas de IA · Ejercicios de equipo rojo estructurados · Seguimiento posterior al despliegueAplicado aquí
- 02Ciberseguridad y criptografíaModelización de amenazas de las aplicaciones de IA · Integridad de la cadena de suministro de modelos · Pruebas de defensa contra la inyecciónAplicado aquí
- 03Inteligencia artificial y aprendizaje automáticoAnálisis del cambio de distribución · Ajuste fino para un uso específico · Métodos de interpretabilidadAplicado aquí
- 04Probabilidad y estadísticaIncertidumbre de las puntuaciones de referencia · Evaluación de la calibración · Muestreo de auditoríaAplicado aquí
Where this page’s references lead.
Los organismos y documentos con los que se lee esta página. El estado y las ediciones cambian; cada enlace lleva al organismo emisor.
Preguntas y respuestas.
What is an AI agent?
What is prompt injection?
How can agent risk be reduced?
Should agents ever act without human approval?
Can agents be evaluated before deployment?
Does the platform build agents?
Aporte evidencia desde esta página.
serie 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.
- Correo electrónicoPasarela GW-GCRIEnvíelo desde esta páginaEvidencia, una corrección, un caso, un método, una revisión, una traducción o un aporte a un perfil. Llega etiquetado con esta página, de modo que lo recibe el editor responsable.
- ZenodoDepósito abiertoDeposítelo y envíelo a la comunidadPublique su propio registro en Zenodo con su ORCID iD y luego envíelo a la comunidad de Reports allí. Los curadores lo revisan; el DOI sigue siendo suyo.
- serie ReportsGCRIVer todas las formas de contribuirLos tipos de contribución, cómo se revisa cada uno, qué crédito recibe, la licencia y qué nunca se acepta.
No envíe nunca datos personales, detalles sensibles de infraestructura ni nada que no tenga derecho a compartir. Un aporte se revisa antes de su uso; enviarlo no confiere autoría ni derechos editoriales, y el patrocinio nunca compra un hallazgo.
Aporte un método, un conjunto de datos o una pregunta.
La información y las consultas están abiertas hoy. Una consulta recibe una referencia e indica el siguiente paso; las solicitudes, pagos, asignaciones de cómputo y registros solo se abren cuando existen su equipo, sus condiciones y la capacidad de revisión.
Documentos jurídicos.
Los mapas, datos y cifras de este sitio se muestran como contexto e información general. No son consejos, pronósticos ni avisos; los límites de la información se exponen en los Descargos de responsabilidad y los Avisos sobre datos.