{"id":946,"date":"2026-01-13T08:30:40","date_gmt":"2026-01-13T08:30:40","guid":{"rendered":"https:\/\/naaia.ai\/news\/ai-agents-risks-sectoral-uses-governance\/"},"modified":"2026-06-02T14:03:18","modified_gmt":"2026-06-02T14:03:18","slug":"ai-agents-risks-sectoral-uses-governance","status":"publish","type":"post","link":"https:\/\/naaia.ai\/en\/ai-agents-risks-sectoral-uses-governance\/","title":{"rendered":"AI\u00a0Agents:\u00a0sectoral\u00a0uses,\u00a0opportunities, and risks to be\u00a0managed\u00a0"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Since&nbsp;2023, AI agents have&nbsp;moved&nbsp;from&nbsp;the&nbsp;experimental&nbsp;stage to&nbsp;operational&nbsp;uses&nbsp;across&nbsp;many&nbsp;sectors: finance,&nbsp;healthcare,&nbsp;industry,&nbsp;human&nbsp;resources, and public services. Capable of acting&nbsp;autonomously&nbsp;or&nbsp;semi-autonomously,&nbsp;these&nbsp;agents promise&nbsp;significant&nbsp;gains in&nbsp;productivity&nbsp;and performance.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">However,&nbsp;this&nbsp;increased&nbsp;autonomy&nbsp;comes&nbsp;with&nbsp;legal,&nbsp;ethical,&nbsp;operational, and&nbsp;cybersecurity&nbsp;risks,&nbsp;making&nbsp;a&nbsp;structured&nbsp;approach&nbsp;to&nbsp;their&nbsp;<a href=\"https:\/\/naaia.ai\/ai-agent-governance-responsibility\/\">governance<\/a>&nbsp;and management essential.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">1.&nbsp;What&nbsp;is an AI&nbsp;agent?&nbsp;Definition&nbsp;and&nbsp;recent&nbsp;evolution&nbsp;<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/naaia.ai\/governance-artificial-intelligence-ai-agents\/\">AI agents<\/a>, in&nbsp;their&nbsp;common&nbsp;understanding,&nbsp;refer&nbsp;to AI system software&nbsp;with&nbsp;specific&nbsp;characteristics:&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>They&nbsp;are&nbsp;based&nbsp;on an <a href=\"https:\/\/naaia.ai\/ai-act-obligations-general-purpose-ai-models-gpai\/\">AI model<\/a>&nbsp;pursuing&nbsp;a&nbsp;defined&nbsp;or&nbsp;undefined&nbsp;objective,&nbsp;which&nbsp;has not&nbsp;undergone&nbsp;additional&nbsp;development&nbsp;or&nbsp;significant&nbsp;modification.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>They&nbsp;are accessible&nbsp;through&nbsp;a studio in&nbsp;which&nbsp;users&nbsp;can&nbsp;edit&nbsp;their&nbsp;parameters.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>They&nbsp;are&nbsp;configured&nbsp;with&nbsp;the&nbsp;aim&nbsp;of&nbsp;automating&nbsp;a&nbsp;complex,&nbsp;contextualized&nbsp;task,&nbsp;making&nbsp;decisions, and&nbsp;executing&nbsp;actions&nbsp;without&nbsp;necessarily&nbsp;requiring&nbsp;human&nbsp;intervention.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">AI agents relate to the concept of&nbsp;<a href=\"https:\/\/naaia.ai\/ai-agents-vs-agentic-ai-differences-definition-2025\/\">agency<\/a>,&nbsp;that&nbsp;is, a&nbsp;system\u2019s&nbsp;ability&nbsp;to:&nbsp;<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Act&nbsp;autonomously,&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Initiate&nbsp;actions,&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Plan&nbsp;sequences,&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Adapt&nbsp;to&nbsp;changing&nbsp;contexts,&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Pursue&nbsp;high-level&nbsp;objectives&nbsp;without&nbsp;continuous&nbsp;human&nbsp;supervision.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><em><strong>Example:<\/strong>&nbsp;an AI agent&nbsp;may&nbsp;be&nbsp;an&nbsp;automated&nbsp;assistant&nbsp;tasked&nbsp;with&nbsp;sorting&nbsp;incoming&nbsp;emails. It&nbsp;analyzes&nbsp;each&nbsp;message, identifies&nbsp;its&nbsp;category&nbsp;(sales, support, urgent), and&nbsp;then&nbsp;applies&nbsp;the&nbsp;appropriate&nbsp;action,&nbsp;such&nbsp;as&nbsp;archiving&nbsp;or ticket&nbsp;creation. It&nbsp;thus&nbsp;performs&nbsp;a&nbsp;specific&nbsp;task,&nbsp;defined&nbsp;in&nbsp;advance,&nbsp;without&nbsp;exceeding&nbsp;the scope of&nbsp;this&nbsp;role.&nbsp;<\/em><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The&nbsp;emergence&nbsp;of&nbsp;frameworks&nbsp;such&nbsp;as&nbsp;AutoGPT&nbsp;(2023),&nbsp;LangGraph&nbsp;(2024), or agents&nbsp;integrated&nbsp;into&nbsp;cloud suites (Microsoft&nbsp;Copilot, Google&nbsp;Agentspace) has&nbsp;accelerated&nbsp;the adoption of AI agents in real&nbsp;professional&nbsp;environments.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">2. Uses of AI agents by&nbsp;sector&nbsp;<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.1 Finance and&nbsp;insurance&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The&nbsp;financial&nbsp;sector&nbsp;is&nbsp;among&nbsp;the first to have&nbsp;integrated&nbsp;AI agents, due to the&nbsp;growing&nbsp;complexity&nbsp;of&nbsp;operations, the&nbsp;increase&nbsp;in data volumes, and the multiplication of compliance&nbsp;requirements.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AI agent use cases<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Risk&nbsp;analysis&nbsp;agents:<\/strong>&nbsp;AI agents&nbsp;continuously&nbsp;assess&nbsp;portfolios,&nbsp;detect&nbsp;anomalies, and&nbsp;adjust&nbsp;risk&nbsp;scores&nbsp;based&nbsp;on&nbsp;internal&nbsp;and&nbsp;external&nbsp;data.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Compliance&nbsp;agents:<\/strong>&nbsp;they&nbsp;ensure&nbsp;continuous&nbsp;monitoring of transactions (AML \/ KYC),&nbsp;prioritize&nbsp;alerts, and&nbsp;prepare&nbsp;compliance files for&nbsp;human&nbsp;validation.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Autonomous&nbsp;algorithmic&nbsp;trading:<\/strong>&nbsp;some&nbsp;agents&nbsp;automatically&nbsp;execute&nbsp;orders&nbsp;according&nbsp;to&nbsp;predefined&nbsp; strategies,&nbsp;based&nbsp;on&nbsp;market&nbsp;conditions and&nbsp;risk&nbsp;constraints.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Associated&nbsp;risks&nbsp;and&nbsp;specific&nbsp;challenges of AI agents<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The use of AI agents in&nbsp;financial&nbsp;activities&nbsp;raises&nbsp;major&nbsp;governance&nbsp;challenges:&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Lack&nbsp;of&nbsp;decision&nbsp;explainability:<\/strong>&nbsp;decisions&nbsp;made or&nbsp;recommended&nbsp;by AI agents&nbsp;may&nbsp;be&nbsp;opaque and&nbsp;therefore&nbsp;difficult&nbsp;to&nbsp;explain,&nbsp;which&nbsp;is&nbsp;problematic&nbsp;in light of&nbsp;regulatory&nbsp;requirements&nbsp;(auditability,&nbsp;traceability, justification of&nbsp;decisions).&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong><a href=\"https:\/\/naaia.ai\/ai-bias-governance\/\">Bias<\/a>&nbsp;and indirect discrimination:<\/strong>&nbsp; underlying&nbsp;models&nbsp;may&nbsp;reproduce&nbsp;or&nbsp;amplify&nbsp;biases&nbsp;present&nbsp;in&nbsp;historical&nbsp;data,&nbsp;leading&nbsp;to&nbsp;unfair&nbsp;risk&nbsp;assessments&nbsp;for certain&nbsp;customer&nbsp;profiles.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Legal and&nbsp;financial&nbsp;liability:<\/strong>&nbsp;in the&nbsp;event&nbsp;of&nbsp;financial&nbsp;loss,&nbsp;undetected&nbsp;fraudulent&nbsp;transactions, or&nbsp;erroneous&nbsp;decisions&nbsp;made by an&nbsp;autonomous&nbsp;agent, the question of&nbsp;human,&nbsp;organizational, or&nbsp;technological&nbsp;responsibility&nbsp;remains&nbsp;complex&nbsp;and&nbsp;requires&nbsp;a&nbsp;clear&nbsp;supervisory&nbsp;framework.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.2 Healthcare and life sciences&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The&nbsp;<a href=\"https:\/\/naaia.ai\/embedded-ai-medical-devices-ai-act-compliance\/\">healthcare&nbsp;and life sciences&nbsp;sectors<\/a>&nbsp;offer&nbsp;strong&nbsp;potential&nbsp;for the use of AI agents,&nbsp;given&nbsp;the&nbsp;complexity&nbsp;of&nbsp;medical&nbsp;data,&nbsp;pressure on&nbsp;healthcare&nbsp;systems, and&nbsp;growing&nbsp;needs&nbsp;for&nbsp;clinical&nbsp;decision&nbsp;support.&nbsp;These&nbsp;agents must&nbsp;be&nbsp;designed&nbsp;as assistance&nbsp;tools&nbsp;and must not replace&nbsp;healthcare&nbsp;professionals.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AI agent use cases<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Diagnostic support&nbsp;agents:<\/strong>&nbsp;AI agents&nbsp;analyze&nbsp;medical&nbsp;records,&nbsp;laboratory&nbsp;results, and&nbsp;imaging&nbsp;to&nbsp;identify&nbsp;clinical&nbsp;signals&nbsp;and&nbsp;suggest&nbsp;diagnostic&nbsp;paths,&nbsp;notably&nbsp;by&nbsp;comparison&nbsp;with&nbsp;cohorts&nbsp;of&nbsp;similar&nbsp;patients.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Care coordination&nbsp;agents:<\/strong>&nbsp;they&nbsp;automate the&nbsp;scheduling&nbsp;of&nbsp;appointments,&nbsp;examinations, and follow-ups,&nbsp;helping&nbsp;to&nbsp;streamline&nbsp;care&nbsp;delivery&nbsp;and&nbsp;optimize&nbsp;the use of&nbsp;hospital&nbsp;resources.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Clinical&nbsp;research:<\/strong>&nbsp;in life sciences, AI agents explore&nbsp;scientific&nbsp;literature&nbsp;and&nbsp;clinical&nbsp;trial data to&nbsp;identify&nbsp;correlations,&nbsp;formulate&nbsp;hypotheses, and&nbsp;accelerate&nbsp;biomedical&nbsp;research.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Associated&nbsp;risks&nbsp;and&nbsp;specific&nbsp;challenges of AI agents<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The use of AI agents in&nbsp;healthcare&nbsp;and life sciences&nbsp;raises&nbsp;major issues&nbsp;related&nbsp;to&nbsp;security,&nbsp;ethics, and&nbsp;responsibility:&nbsp;<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Protection of&nbsp;health&nbsp;data:<\/strong>&nbsp;AI agents process&nbsp;highly&nbsp;sensitive&nbsp;medical&nbsp;data,&nbsp;increasing&nbsp;the&nbsp;risk&nbsp;of&nbsp;privacy&nbsp;breaches&nbsp;in the&nbsp;event&nbsp;of&nbsp;security&nbsp;failures,&nbsp;poor&nbsp;access&nbsp;governance, or&nbsp;uncontrolled&nbsp;data use.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Risk of&nbsp;medical&nbsp;errors:<\/strong>&nbsp;misinterpretation&nbsp;of data, model&nbsp;biases, or&nbsp;incomplete&nbsp;clinical&nbsp;information&nbsp;may&nbsp;lead to&nbsp;inaccurate&nbsp;recommendations,&nbsp;with&nbsp;potential&nbsp;consequences&nbsp;for&nbsp;quality&nbsp;of care and patient&nbsp;safety.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Excessive&nbsp;reliance&nbsp;on&nbsp;algorithmic&nbsp;recommendations:<\/strong>&nbsp;increased&nbsp;use of AI agents&nbsp;may&nbsp;weaken&nbsp;clinical&nbsp;judgment&nbsp;if not&nbsp;properly&nbsp;supervised,&nbsp;making&nbsp;it&nbsp;essential to&nbsp;implement&nbsp;mechanisms&nbsp;for&nbsp;human&nbsp;oversight,&nbsp;explainability, and&nbsp;clearly&nbsp;defined&nbsp;responsibility.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.3 Human&nbsp;resources&nbsp;and talent management&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Human&nbsp;resources&nbsp;functions&nbsp;represent&nbsp;a prime area of application for AI agents, in a&nbsp;context&nbsp;marked&nbsp;by an&nbsp;increasing&nbsp;number&nbsp;of applications,&nbsp;rapid&nbsp;skills&nbsp;evolution, and the&nbsp;need&nbsp;to&nbsp;better&nbsp;anticipate&nbsp;talent&nbsp;needs.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AI agent use cases<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Candidate&nbsp;pre-screening&nbsp;agents:<\/strong>&nbsp;AI agents&nbsp;analyze&nbsp;CVs, cover&nbsp;letters, and&nbsp;professional&nbsp;profiles to&nbsp;identify&nbsp;candidates best&nbsp;matching&nbsp;defined&nbsp;criteria,&nbsp;while&nbsp;prioritizing&nbsp;profiles for&nbsp;recruiters&nbsp;to&nbsp;review.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Automated&nbsp;onboarding&nbsp;agents:<\/strong>&nbsp;they&nbsp;support new&nbsp;employees&nbsp;during&nbsp;their&nbsp;integration&nbsp;by&nbsp;automating&nbsp;certain administrative&nbsp;steps,&nbsp;delivering&nbsp;personalized&nbsp;content, and&nbsp;facilitating&nbsp;role&nbsp;onboarding.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Skills&nbsp;management and&nbsp;internal&nbsp;mobility&nbsp;agents:<\/strong>&nbsp;these&nbsp;agents cross-reference&nbsp;HR data,&nbsp;internal&nbsp;evaluations, and business&nbsp;needs&nbsp;to&nbsp;identify&nbsp;skills&nbsp;gaps,&nbsp;recommend&nbsp;training, and propose&nbsp;mobility&nbsp;opportunities.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Associated&nbsp;risks&nbsp;and&nbsp;specific&nbsp;challenges of AI agents<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The use of AI agents in HR&nbsp;raises&nbsp;significant&nbsp;challenges&nbsp;related&nbsp;to&nbsp;fairness, compliance, and&nbsp;governance:&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Risk of indirect&nbsp;discrimination:<\/strong>&nbsp;models&nbsp;may&nbsp;reproduce&nbsp;or&nbsp;amplify&nbsp;biases&nbsp;related&nbsp;to&nbsp;age,&nbsp;gender, or&nbsp;origin&nbsp;due to non-neutral&nbsp;historical&nbsp;data.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Protection of&nbsp;personal&nbsp;data and&nbsp;regulatory&nbsp;compliance:<\/strong>&nbsp;AI agents process sensitive data&nbsp;covered&nbsp;by the GDPR,&nbsp;requiring&nbsp;high standards of&nbsp;transparency, data&nbsp;minimization, and information for data&nbsp;subjects.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Human control over&nbsp;decisions:<\/strong>&nbsp;excessive&nbsp;reliance&nbsp;on automation&nbsp;may&nbsp;reduce&nbsp;human&nbsp;involvement&nbsp;in&nbsp;structuring&nbsp;decisions&nbsp;such&nbsp;as&nbsp;recruitment&nbsp;or&nbsp;career&nbsp; development,&nbsp;making&nbsp;clear&nbsp;mechanisms&nbsp;for supervision and&nbsp;accountability&nbsp;essential.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.4&nbsp;Industry,&nbsp;supply&nbsp;chain, and&nbsp;logistics&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Industry,&nbsp;supply&nbsp;chain, and&nbsp;logistics&nbsp;are key areas for the application of AI agents, due to the&nbsp;complexity&nbsp;of value&nbsp;chains, the&nbsp;multiplicity&nbsp;of stakeholders, and the&nbsp;need&nbsp;to&nbsp;continuously&nbsp;optimize&nbsp;production and&nbsp;supply&nbsp;flows.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AI agent use cases<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Predictive&nbsp;maintenance&nbsp;agents:<\/strong>&nbsp;AI agents&nbsp;continuously&nbsp;analyze&nbsp;data&nbsp;from&nbsp;industrial&nbsp;sensors&nbsp;and maintenance histories to&nbsp;anticipate&nbsp;failures, plan interventions, and&nbsp;reduce&nbsp;unplanned&nbsp;downtime.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Supply&nbsp;chain&nbsp;optimization&nbsp;agents:<\/strong>&nbsp;they&nbsp;combine&nbsp;demand&nbsp;data, production&nbsp;capacities,&nbsp;inventory&nbsp;levels, and&nbsp;logistical&nbsp;constraints&nbsp;to&nbsp;adjust&nbsp;flows,&nbsp;limit&nbsp;shortages, and&nbsp;reduce&nbsp;costs.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Real-time production planning&nbsp;agents:<\/strong>&nbsp;these&nbsp;agents&nbsp;adapt&nbsp;production&nbsp;schedules&nbsp;based&nbsp;on&nbsp;operational&nbsp;disruptions,&nbsp;demand&nbsp;fluctuations, or&nbsp;external&nbsp;constraints.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Associated&nbsp;risks&nbsp;and&nbsp;specific&nbsp;challenges of AI agents<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The use of AI agents in&nbsp;industry&nbsp;and&nbsp;logistics&nbsp;raises&nbsp;significant&nbsp;challenges&nbsp;related&nbsp;to&nbsp;reliability,&nbsp;security, and&nbsp;governance:&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Cascading&nbsp;effects&nbsp;of&nbsp;automated&nbsp;decisions:<\/strong>&nbsp;a configuration or&nbsp;decision&nbsp;error&nbsp;may&nbsp;quickly&nbsp;propagate&nbsp;across&nbsp;the&nbsp;entire&nbsp;value&nbsp;chain,&nbsp;affecting&nbsp;production, inventories, and&nbsp;deliveries.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Dependence&nbsp;on&nbsp;external&nbsp;data:<\/strong>&nbsp;the&nbsp;quality&nbsp;of AI agents\u2019&nbsp;decisions&nbsp;relies on data&nbsp;that&nbsp;may&nbsp;be&nbsp;incomplete&nbsp;or&nbsp;unreliable,&nbsp;potentially&nbsp;weakening&nbsp;operational&nbsp;trade-offs.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Cyber&nbsp;vulnerabilities&nbsp;of&nbsp;industrial&nbsp;systems:<\/strong>&nbsp;integrating&nbsp;AI agents&nbsp;into&nbsp;critical&nbsp;industrial&nbsp;environments&nbsp;increases&nbsp;the&nbsp;attack&nbsp;surface and&nbsp;requires&nbsp;strengthened&nbsp;cybersecurity&nbsp;and&nbsp;access&nbsp;control&nbsp;measures.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.5 Public&nbsp;sector&nbsp;and citizen services&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The public&nbsp;sector&nbsp;and&nbsp;citizen&nbsp;services&nbsp;represent&nbsp;a&nbsp;growing&nbsp;field&nbsp;of application for AI agents, in a&nbsp;context&nbsp;marked&nbsp;by&nbsp;increasing&nbsp;volumes of administrative&nbsp;requests, the&nbsp;search&nbsp;for&nbsp;greater&nbsp;efficiency, and the&nbsp;need&nbsp;to&nbsp;ensure&nbsp;equitable&nbsp;access&nbsp;to public services,&nbsp;while&nbsp;providing&nbsp;reinforced&nbsp;oversight&nbsp;given&nbsp;<a href=\"https:\/\/naaia.ai\/high-risk-ai-act-list-eu-ai-regulation-compliance\/\">fundamental&nbsp;rights&nbsp;concerns<\/a>.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AI agent use cases<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>User orientation&nbsp;agents:<\/strong>&nbsp;AI agents assist&nbsp;citizens&nbsp;with&nbsp;administrative&nbsp;procedures&nbsp;by&nbsp;directing&nbsp;them&nbsp;to the&nbsp;appropriate&nbsp;services,&nbsp;automating&nbsp;certain&nbsp;responses, and&nbsp;facilitating&nbsp;access&nbsp;to public information.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Administrative&nbsp;decision-support&nbsp;agents:<\/strong>&nbsp;they&nbsp;analyze&nbsp;complex&nbsp;files to&nbsp;prioritize&nbsp;processing&nbsp;or&nbsp;formulate&nbsp;recommendations&nbsp;for public&nbsp;officials,&nbsp;particularly&nbsp;in social&nbsp;aid&nbsp;or&nbsp;resource&nbsp;allocation.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Social or&nbsp;tax&nbsp;fraud&nbsp;detection&nbsp;agents:<\/strong>&nbsp;these&nbsp;agents cross-reference&nbsp;large volumes of administrative data to&nbsp;identify&nbsp;inconsistencies&nbsp;or&nbsp;atypical&nbsp;behaviors&nbsp;likely&nbsp;to&nbsp;reveal&nbsp;fraudulent&nbsp;situations.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Associated&nbsp;risks&nbsp;and&nbsp;specific&nbsp;challenges of AI agents<\/strong>&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The use of AI agents in the public&nbsp;sector&nbsp;raises&nbsp;particularly&nbsp;sensitive issues&nbsp;related&nbsp;to&nbsp;transparency,&nbsp;fairness, and&nbsp;democratic&nbsp;accountability:&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Infringement&nbsp;of&nbsp;fundamental&nbsp;rights:<\/strong>&nbsp;poorly&nbsp;supervised&nbsp;automated&nbsp;decisions&nbsp;may&nbsp;affect&nbsp;access&nbsp;to&nbsp;rights, social&nbsp;benefits, or essential services.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Opacity&nbsp;of&nbsp;decision&nbsp;criteria:<\/strong>&nbsp;the&nbsp;lack&nbsp;of&nbsp;clear&nbsp;explainability&nbsp;of&nbsp;algorithmic&nbsp;recommendations&nbsp;complicates&nbsp; their&nbsp;understanding&nbsp;by&nbsp;both&nbsp;public&nbsp;officials&nbsp;and&nbsp;citizens.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Insufficient&nbsp;contestability&nbsp;of&nbsp;decisions:<\/strong>&nbsp;the&nbsp;lack&nbsp;of&nbsp;appeal&nbsp;mechanisms&nbsp;and&nbsp;human&nbsp;control&nbsp;may&nbsp;limit&nbsp;users\u2019&nbsp;ability&nbsp;to challenge an&nbsp;automated&nbsp;decision,&nbsp;making&nbsp;the&nbsp;implementation&nbsp;of&nbsp;appropriate&nbsp;procedural&nbsp;safeguards&nbsp;essential.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">3. Major cross-cutting risks of AI agents&nbsp;<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Beyond&nbsp;sector-specific&nbsp;issues, the&nbsp;deployment&nbsp;of AI agents&nbsp;raises&nbsp;cross-cutting&nbsp;risks&nbsp;common&nbsp;to all&nbsp;organizations.&nbsp;These&nbsp;risks&nbsp;concern&nbsp;<a href=\"https:\/\/naaia.ai\/cost-of-non-compliance-ia-act\/\">legal,&nbsp;ethical,&nbsp;operational, and&nbsp;cybersecurity&nbsp;dimensions<\/a> and call for a global AI&nbsp;governance&nbsp;approach.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.1 Legal and&nbsp;Regulatory&nbsp;Risks&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The&nbsp;growing&nbsp;autonomy&nbsp;of AI agents exposes&nbsp;organizations&nbsp;to&nbsp;regulatory&nbsp;non-compliance&nbsp;risks,&nbsp;particularly&nbsp;when&nbsp;these&nbsp;systems&nbsp;contribute&nbsp;to&nbsp;decisions&nbsp;with&nbsp;legal&nbsp;or&nbsp;significant&nbsp;effects&nbsp;on&nbsp;individuals.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Non-compliance&nbsp;with&nbsp;the&nbsp;GDPR:<\/strong>&nbsp;Article 22 of the GDPR&nbsp;strictly&nbsp;regulates&nbsp;fully&nbsp;automated&nbsp;decisions&nbsp;producing&nbsp;legal&nbsp;effects. The use of AI agents&nbsp;without&nbsp;human&nbsp;oversight&nbsp;mechanisms, information to data&nbsp;subjects, or avenues for recourse&nbsp;may&nbsp;constitute&nbsp;a direct violation of the&nbsp;European&nbsp;framework.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Exposure&nbsp;to&nbsp;emerging&nbsp;regulations:<\/strong>&nbsp;the&nbsp;gradual&nbsp;entry&nbsp;into&nbsp;force of the&nbsp;European&nbsp;AI&nbsp;Act, as&nbsp;well&nbsp;as the adoption of AI&nbsp;framework&nbsp;laws&nbsp;in Asia (South&nbsp;Korea, Taiwan, Japan), imposes new obligations&nbsp;regarding&nbsp;risk&nbsp;classification,&nbsp;transparency, and&nbsp;governance.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Uncertain&nbsp;legal&nbsp;liability:<\/strong>&nbsp;in the&nbsp;event&nbsp;of&nbsp;harm&nbsp;caused&nbsp;by an&nbsp;autonomous&nbsp;decision, the allocation of&nbsp;responsibility&nbsp;among&nbsp;the&nbsp;organization,&nbsp;human&nbsp;teams,&nbsp;technology&nbsp;providers, and the AI agent&nbsp;itself&nbsp;remains&nbsp;legally&nbsp;complex.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.2&nbsp;Ethical&nbsp;Risks&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">AI agents&nbsp;also&nbsp;raise&nbsp;structural&nbsp;ethical&nbsp;issues&nbsp;linked&nbsp;to&nbsp;their&nbsp;ability&nbsp;to influence,&nbsp;recommend, or automate sensitive&nbsp;decisions.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Algorithmic&nbsp;bias:<\/strong>&nbsp;agents&nbsp;may&nbsp;reproduce&nbsp;or&nbsp;amplify&nbsp;biases&nbsp;present&nbsp;in training data,&nbsp;leading&nbsp;to indirect discrimination or&nbsp;inequitable&nbsp;treatment.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Weakening&nbsp;of&nbsp;human&nbsp;autonomy:<\/strong>&nbsp;excessive&nbsp;reliance&nbsp;on AI agent&nbsp;recommendations&nbsp;may&nbsp;reduce&nbsp;individuals\u2019&nbsp;ability&nbsp;to&nbsp;exercise&nbsp;critical&nbsp;judgment,&nbsp;particularly&nbsp;in&nbsp;complex&nbsp;decision-making&nbsp;contexts.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Lack&nbsp;of&nbsp;transparency&nbsp;and&nbsp;explainability:<\/strong>&nbsp;the&nbsp;opacity&nbsp;of certain&nbsp;models&nbsp;makes&nbsp;it&nbsp;difficult&nbsp;to&nbsp;understand&nbsp;decision-making&nbsp;logic,&nbsp;undermining&nbsp;user and stakeholder trust.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.3&nbsp;Operational&nbsp;and&nbsp;Cybersecurity&nbsp;Risks&nbsp;<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">From&nbsp;an&nbsp;operational&nbsp;standpoint, AI agents&nbsp;introduce&nbsp;new&nbsp;vectors&nbsp;of&nbsp;technical&nbsp;and&nbsp;organizational&nbsp;risk.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Misconfiguration&nbsp;or&nbsp;misuse&nbsp;of&nbsp;agents:<\/strong>&nbsp;poorly&nbsp;configured&nbsp;or&nbsp;insufficiently&nbsp;controlled&nbsp;agents&nbsp;may&nbsp;produce&nbsp; erroneous&nbsp;decisions&nbsp;or&nbsp;be&nbsp;exploited&nbsp;for&nbsp;malicious&nbsp;purposes.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Excessive&nbsp;access&nbsp;to&nbsp;internal&nbsp;systems:<\/strong>&nbsp;AI agents&nbsp;often&nbsp;require&nbsp;broad&nbsp;access&nbsp;to&nbsp;databases&nbsp;or&nbsp;critical&nbsp;systems,&nbsp;increasing&nbsp;the&nbsp;attack&nbsp;surface in the&nbsp;event&nbsp;of compromise.&nbsp;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Difficulty&nbsp;of ex post&nbsp;auditing:<\/strong>&nbsp;the&nbsp;autonomous&nbsp;chaining&nbsp;of&nbsp;decisions&nbsp;and actions&nbsp;complicates&nbsp;traceability&nbsp;and&nbsp;auditing&nbsp;of agent&nbsp;behavior,&nbsp;particularly&nbsp;in the&nbsp;event&nbsp;of incidents or disputes.&nbsp;<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In the face of the&nbsp;growing&nbsp;power of AI agents, <a href=\"https:\/\/naaia.ai\/aims-definition-governance-compliance-ai\/\">AI management<\/a>&nbsp;is&nbsp;becoming&nbsp;a&nbsp;strategic&nbsp;lever for&nbsp;organizations.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Manage\u00a0your\u00a0AI agents\u00a0with\u00a0confidence\u00a0<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">AI agents are&nbsp;already&nbsp;transforming&nbsp;your&nbsp;business&nbsp;processes. The question&nbsp;is&nbsp;no longer&nbsp;whether&nbsp;they&nbsp;should&nbsp;be&nbsp;used, but how to&nbsp;deploy, supervise, and&nbsp;govern&nbsp;them&nbsp;responsibly.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Naaia&nbsp;supports&nbsp;organizations&nbsp;in AI&nbsp;management:&nbsp;from&nbsp;AI agent&nbsp;inventory&nbsp;to&nbsp;risk&nbsp;management,&nbsp;regulatory&nbsp;compliance, and&nbsp;operational&nbsp;steering.&nbsp;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"> <a href=\"https:\/\/naaia.ai\/request-demo\/\"> Discover&nbsp;our&nbsp;AI management platform and explore solutions to&nbsp;better&nbsp;manage&nbsp;and&nbsp;oversee&nbsp;your&nbsp;AI agents&nbsp;today,&nbsp;while&nbsp;anticipating&nbsp;upcoming&nbsp;regulatory&nbsp;developments.<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Since&nbsp;2023, AI agents have&nbsp;moved&nbsp;from&nbsp;the&nbsp;experimental&nbsp;stage to&nbsp;operational&nbsp;uses&nbsp;across&nbsp;many&nbsp;sectors: finance,&nbsp;healthcare,&nbsp;industry,&nbsp;human&nbsp;resources, and public services. Capable of acting&nbsp;autonomously&nbsp;or&nbsp;semi-autonomously,&nbsp;these&nbsp;agents promise&nbsp;significant&nbsp;gains in&nbsp;productivity&nbsp;and performance.&nbsp; However,&nbsp;this&nbsp;increased&nbsp;autonomy&nbsp;comes&nbsp;with&nbsp;legal,&nbsp;ethical,&nbsp;operational, and&nbsp;cybersecurity&nbsp;risks,&nbsp;making&nbsp;a&nbsp;structured&nbsp;approach&nbsp;to&nbsp;their&nbsp;governance&nbsp;and management essential.&nbsp; 1.&nbsp;What&nbsp;is an AI&nbsp;agent?&nbsp;Definition&nbsp;and&nbsp;recent&nbsp;evolution&nbsp; AI agents, in&nbsp;their&nbsp;common&nbsp;understanding,&nbsp;refer&nbsp;to AI system software&nbsp;with&nbsp;specific&nbsp;characteristics:&nbsp; AI agents&hellip; <a href=\"https:\/\/naaia.ai\/en\/ai-agents-risks-sectoral-uses-governance\/\">Lire la suite<\/a><\/p>\n","protected":false},"author":2,"featured_media":944,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[46],"tags":[],"class_list":["post-946","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-governance-blog"],"_links":{"self":[{"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/posts\/946","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/comments?post=946"}],"version-history":[{"count":1,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/posts\/946\/revisions"}],"predecessor-version":[{"id":2262,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/posts\/946\/revisions\/2262"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/media\/944"}],"wp:attachment":[{"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/media?parent=946"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/categories?post=946"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/naaia.ai\/en\/wp-json\/wp\/v2\/tags?post=946"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}