Inria Releases Policy Paper on Agentic AI, “Agentic AI: Deployment, Adoption and Impacts”, and Presents Nine Key Recommendations

Generative AI has recently evolved from models that primarily produced text, images, or code to systems capable of making autonomous decisions and acting proactively within human-defined boundaries. Unlike traditional generative AI models, agentic AI systems can analyze and respond to signals from their environment, plan actions, execute them, and reflect on the outcomes, often operating continuously with varying degrees of autonomy and adaptive behavior.

The paper identifies key developments to anticipate and the actions needed to address them. It was prepared by an interdisciplinary working group from Inria's Program Agency through an iterative process. The initiative brought together a wide range of stakeholders—including representatives from industry, academia, and the policymaking community—to address emerging challenges and provide practical, forward-looking analysis for decision-makers.

As part of the process, the team conducted industry surveys and organized five workshops with more than 80 participants from academia, the public sector, and private organizations across all G7 countries to gather expert feedback and diverse perspectives.

The paper addresses key topics such as traceability and explainability; safety, control, protection, and harm prevention; energy consumption; and the impact on workers, among other critical issues.

It is structured around four thematic axis:

1. Technical concepts and paradigms

2. Conditions for the successful deployment of AI

3. Regulatory frameworks (with particular emphasis on the European Union context)

4. Recommendations

In this final axis, the paper outlines nine key recommendations to ensure the successful deployment, adoption, and impact of agentic AI, ranked by level of urgency:

1. Ensure strategic autonomy: Diversify suppliers and develop national agentic AI solutions to reduce strategic dependencies and ensure flexibility across the entire technology value chain.

2. Establish a clear and operational definition of agentic AI: Develop a shared, evolving technical definition that can serve as a regulatory foundation, avoiding ambiguity while guiding policymaking, risk classification, and compliance requirements.

3. Foster a fertile ground for Agentic AI research: Implement agile funding mechanisms and facilitate access to advanced tools, while promoting interdisciplinary research to address identified scientific gaps.

4. Ensure legal security for users and developers: Review and adapt existing legal frameworks (such as the EU AI Act) to address the specific characteristics of agentic AI, clearly defining the responsibilities of developers, deployers, and users.

5. Adopt an impact-centered governance approach: Build governance frameworks based on the anticipated social, labor, and environmental impacts of agentic AI, integrating interdisciplinary perspectives while balancing precaution with innovation.

6. Define common ground to compare energy consumption: Develop an energy rating system based on shared benchmarks to provide transparent information on the energy footprint of AI models and multi-agent architectures, while protecting proprietary information and trade secrets.

7. Address cybersecurity risks specific to Agentic AI: Assess the new security threats arising from the open-loop interaction of agentic AI systems with their environments and promote secure architectures for identity and access management.

8. Anticipate workforce transformation and cognitive impacts: Investigate the effects of agentic AI on the workforce, including skill erosion and "cognitive offloading," while integrating insights from the social sciences into adaptation policies.

9. Strengthen public awareness and education on Agentic AI: Promote accessible AI literacy for the general public, clearly distinguishing agentic AI from other technological tools and raising a

• Read the G7 Ministerial Declaration on Digital & Technology (May 2026): G7 Ministerial Declaration on Digital & Technology

• Learn more about the commitments adopted by the G7 to advance a responsible digital transformation under the French Presidency: explore the adopted priorities and related documents here (available in French and English).

• Learn more about the Hiroshima AI Process: discover its key documents here.