Artificial Intelligence in Parliaments - GenAI, Trust, Sovereignty & Innovation

Presented in Brussels at the Flemish Parliament’s reflection day on AI and democracy on 16 October 2025. Full title of the presenation: Artificial Intelligence in Parliaments - Generative AI, Public Trust, Data Sovereignty & Legislative Innovation

1. Artificial Intelligence in Parliaments
Generative AI, Public Trust, Data Sovereignty & Legislative Innovation
Brussels, 16 October 2025
Dr. Fotios Fitsilis
Head of Department, Scientific Service, Hellenic Parliament
Professor of AI & Parliamentary Governance, Universidad Austral

2. Structure
Part I: Introduction, GenAI, Public Trust andData Sovereignt
Part II: AI in Legislation
Part III: AI Tools for Parliamentarians
Part IV: Strategic Integration of AI into the Parliamentary Workspace
2

3. Part I
Public Trust & Data Sovereignty
3

4. From Paper-Based to Data-Driven Institutions
Historical Context
●Centuries of paper-based processes
●Digital transformation waves
●AI as the next paradigm shift
Current Reality
●Dual role: legislators and users
●Parliaments both regulating and adopting AI
●Surveys showing the evolution of AI use cases
4
2022: 39 use cases (Fitsilis & de Almeda, 2024)
2024: 65 use cases (IPU, 2024)

5. Categories of Parliaments
1. Advanced AI Adopters
●Already implementing AI solutions
●Sophisticated use cases
●Setting benchmarks
2. Experimental Explorers
●Pilot projects underway
●Testing AI applications
●Building capacity
3. Foundation Builders
●Developing basic IT infrastructure
●Establishing data governance
●Prerequisites for AI adoption
5
Foundation
Builders
Experimental
Explorers
Advanced AI
Adopters

6. 6

7. Real-World EU Parliamentary AI Applications
7

8. TheEuropean Perspective
The Challenge
●Corporate-controlled AI
●Profit-driven models vs.
public interest
●Dependencies on foreign technology
EU AI Factories Initiative
●Goal: European AI independence
●13 AI Factories in the making
Parliamentary Implications
●Sovereign infrastructure
●Data governance
●Control over legislative information
8

9. 95% of GenAI Pilots at Companies Are Failing
Key Facts
• MIT study: 150 exec interviews, 350
employee surveys, 300 public deployments
• Only 5% generate measurable return or
scale beyond pilot
• Huge corporate investments in GenAI, but
limited profit impact
Causes & Insights
• Messy data, weak integration, poor
alignment with business processes
• Overinvestment in sales/marketing pilots
instead of backend automation
• Success requires external partners, strong
data infrastructure, workflow-first design
Blueprint for the 5% Success Cases
• Align AI efforts to core business processes and KPIs
• Invest in robust data pipelines, governance, and feedback loops
• Choose focused vendors or external partners when internal capacity is weak
• Start with use cases that have clear ROI (e.g. back-office automation)
© MIT NANDA -GenAI Divide: State of AI in Business 2025

10. Key Considerations for LLM-based apps
●Three main options
1.Use existing closed, pre-trained Models from cloud providers using an API
2.Deploy and fine-tune new LLMs on the cloud
3.Deploy and fine-tune new LLMs on-premises
●Factors to consider
○Data privacy and compliance
○Cost
○Customization needs
○Scalability
○Ease of use and deployment
○Transparency
10

11. ThePublic TrustImperative
Why Trust Matters in Parliaments
●Democratic legitimacy depends on public confidence
●AI decisions must be transparent and accountable
●Mistakes can undermine institutional credibility
Trust Challenges
●Black box algorithms vs. democratic transparency
●Data privacy concerns
●Algorithmic bias and fairness
●Accountability in AI-assisted decisions
Building Trust
●Largely explainable AI systems
●Human oversight mechanisms
●Clear guidelines and policies 11

12. Part II
12

13. AI-Enhanced Legislative Drafting
Current Challenges(indicative)
●Complex legal language
●Consistency
●Cross-references and dependencies
●Time pressure
EU Project: AI-based Solutions for
Legislative Drafting
●Cooperation between parliament,
academia and the private sector
●Practical, opentools for EU
institutions
13
Fitsilis & Mikros(2024)

14. AI Apps = Smart Functionalities
Approach
●“Smart Functionalities” principle
●Identification on 34 practical applications
●Practical implementation for LEOS/EdiT
Human-AI Collaboration
●AI suggests, humans decide
●Expert review remains essential
●Augmentation, not replacement
14
Fitsilis & Mikros(2024)

15. Strengthening Parliamentary Oversight
Monitoring Executive Actions
●Automated tracking of government commitments
●Implementation monitoring
●Budget execution analysis
Committee Support
●Research assistance
●Trend identification in hearings
●Report generation
Question Time Enhancement
●Similar question detection
●Answer quality assessment
●Follow-up suggestion generation
15

16. Part III
16

17. AI as Research Assistant
For Parliamentary Research Services
●Accelerated research cycles
●More comprehensive analysis
●Pattern detection in large datasets
●Early warning systems
Efficiency Gains
●More time for strategic thinking
●Better-informed debates
●Higher quality scrutiny
17

18. Risks and Mitigation Strategies
Key Risks
●Algorithmic bias affecting decisions
●Over-reliance on AI outputs
●Security vulnerabilities
●Loss of human judgment
●Public perception of "robot parliament"
Mitigation Strategies
●Comprehensive guidelines and policies
●Training for parliamentarians and staff
●Regular audits and assessments
●Transparency about AI use
●Maintain human decision-making authority
Critical Principle
●New technologies must serve democracy, not distort it
18

19. Developing AI Guidelines for Parliaments
Structure
●40 guidelines
●Grouped in 6 sections
Development
●Hellenic OCR Team
●Partner institutions
●International experts
●Published by Westminster Foundation for Democracy
Approach
●22 parliamentary experts from 16 countries
●Multi-disciplinary and cross-sector development
●Based on scientific frameworks; Corporate principles;
Practical pilot project experiences
19
Fitsilis, von Lucke & De Vrieze (Eds) (2024)

20. 20

21. 21

22. Modernization to Prevent AI-Induced Executive Dominance
The Risk
●Executives may adopt AI faster
●Information asymmetry increases
●Parliamentary oversight weakens
●Democratic balance threatened
Parliamentary Response
●Proactive AI adoption
●Building internal capacity
●Independent AI capabilities
●Maintaining institutional autonomy
22

23. Call to Action
For MPs
●Engage with AI technologies
●Demand proper guidelines
●Participate in training
●Champion responsible adoption
For parliamentary administration
●Develop comprehensive strategies
●Build internal capacity
●Establish governance frameworks
●Foster innovation culture
23
For international parliamentary
community
•Share experiences openly
•Collaborate on standards
•Support capacity building
•Create knowledge networks

24. Thank you!
Questions & Discussion
Prof. Dr. Fotios Fitsilis
Hellenic Parliament | Universidad Austral
For more information: fitsilis.gr| ResearchGate| LinkedIn
Some visual elements were created with the support of design and AI tools, including Canva, ChatGPT and Claude
This presentation reflects the personal views of the authors and does not necessarily represent those of their institutions