Anticipating the Agentic Era: Assessing the Disruptions by AI Agents

November 19th, 2025, 09h30-17h30,
Starling Hotel,
1025 Saint-Sulpice

Introduction

As we enter the agentic era, AI agents are increasingly integrated into various aspects of modern life, performing tasks that range from personal assistance and financial management to complex decision-making in industries. These AI agents, driven by powerful algorithms, are transforming how we interact with technology and each other. However, their proliferation also introduces significant challenges, such as cybersecurity vulnerabilities, economic and business disruptions, ethical dilemmas, and privacy concerns. This conference aims to delve into how AI agents could challenge existing systems and structures while exploring strategies to mitigate these threats effectively. Industry leaders, researchers and academics, and stakeholders will convene to discuss the implications of AI agents and collaborate on building a resilient future that harnesses their potential responsibly.

This event is organized by the Center for Digital Trust (C4DT), EPFL.

Objectives

  • Assess the transformative impact of AI agents on cybersecurity, the economy, business models, governance, regulation, and society.
  • Facilitate expert dialogue on the risks, ethical challenges, and opportunities introduced by agentic systems.
  • Discuss strategies and frameworks for safely and responsibly deploying, governing, and regulating autonomous AI agents.

Discussion points

1. Evolving Cyber Threats in the Age of AI Agents

Explore the unique cybersecurity risks introduced by autonomous AI agents, examining model, code, and prompt-based vulnerabilities as well as new modalities of data exfiltration and privacy breaches. Through expert talks and case studies, participants will discuss strategies and solutions to address the increasingly complex threat landscape created by agentic systems.

2. Evaluation and Benchmarking of AI Agents

Examine the critical importance of robust evaluation and transparent benchmarking for AI agents, discussing methodologies to assess their capabilities, reliability, and trustworthiness. The conference will address how standardized evaluation practices can guide safer deployment, continuous improvement, and informed oversight of agentic systems.

3. AI Agents and the Transformation of the Workforce

Examine the profound impact of AI agents on the job market, discussing how automation is shifting employment patterns and redefining the skills required across industries. The conference aims to address the challenges of workforce displacement and to explore forward-thinking approaches for worker reskilling and policy adaptation.

4. Accelerating Innovation and Transforming Education & Research

Discuss how AI agents are compressing research timelines and transforming educational and research paradigms, examining both the tremendous opportunities for faster knowledge creation and the challenges to established teaching and learning models. The session will explore new frameworks for harnessing these advancements while addressing potential risks and inequities.

5. Governing Algorithms: Accountability in Agentic Systems

Examine the governance challenges that arise as AI agents become increasingly autonomous, exploring frameworks for transparency, oversight, and assigning responsibility for agentic actions. Stakeholders will discuss how to address gaps in accountability and ensure robust mechanisms for remediation in cases of harm or malfunction.

6. Regulatory Frontiers: Crafting Laws for Autonomous Agents

Address the limitations of traditional regulatory approaches in the face of rapidly developing AI agents, exploring innovative legal frameworks and adaptive policy tools. The conference will discuss how best to regulate agentic systems to ensure safety, fairness, and ethical deployment without curtailing beneficial innovation.

7. Navigating the Autonomy-Control Spectrum

Discuss the growing tension between granting AI agents autonomy and retaining essential human oversight, exploring methods to calibrate agentic independence with safety and controllability. The conference will examine best practices and scenarios to address potential risks associated with unchecked or misaligned autonomy.

8. Governance and Safety in Self-Improving Agents

Explore the unique governance challenges and safety concerns posed by AI agents capable of recursive self-improvement. The participants will discuss strategies for maintaining operational control, ensuring transparency in agent behavior, and coordinating multi-stakeholder governance to prevent unintended consequences and safeguard societal interests as these agents evolve.

Agenda

09h00

Registration and welcome coffee

09h30

Welcome and introductory remarks

by David Viollier, C4DT

Part 1 – Tech & Cybersecurity

Begin the conference by exploring the emerging technical challenges and cybersecurity risks posed by autonomous AI agents, focusing on vulnerabilities, attack vectors, and innovative solutions such as autonomous bug fixing to strengthen system resilience and security.

09h35

Talk: AI coding Agent vulnerabilities

by Nils Amiet, Lead Prototyping Engineer, Kudelski Security

10h00

Talk: MCP vulnerability, prompt injections & data exfiltration by AI agents

by Julia Bazinska, Senior Research Engineer Lakera AI

10h25

Talk: Autonomous bug fixing using AI agents, benchmarking

by Mark Müller, Co-Founder and CTO, LogicStar AI

10h50

Coffee Break

Part 2 – Evaluation & Benchmarking

Turn attention to the essential practices of evaluating and benchmarking AI agents, highlighting why transparency, observability, and robust metrics are vital for trustworthy and safe deployment.

11h20

Talk: Building Trust in Agentic Systems: Thomson Reuters’ Approach to Evaluation and Observability

by Lea Strohm, Senior Technical Product Manager, Thomson Reuters

11h45

Panel 1 – AI agent evaluation & benchmarking

“You can’t trust what you can’t observe and evaluate, and you can’t improve what you don’t benchmark” – discussing the need for robust evaluation and transparent benchmarking to ensure the trustworthiness of agentic systems.

Moderated by: [to be confirmed]

Panelists

  • Julia Bazinska, Senior Research Engineer, Lakera AI
  • Prof. Martin Jaggi, Head of Machine Learning and Optimization Laboratory, EPFL
  • Mark Müller, Co-Founder and CTO, LogicStar AI
  • Lea Strohm, Senior Technical Product Manager, Thomson Reuters


12h30


Lunch

Part 3 – Economic/Business and Innovation

Explore the transformative impact of AI agents on the economy, business models, and innovation dynamics, discussing shifts in the labor market, disruptions to established industries, and the acceleration of research and learning processes.

13h30

Talk: Economic impact of AI on job market, extrapolated to AI agents

by Prof. Marcel Salathé, Co-Director, EPFL AI Center

13h55

Talk: Exploring Business Model Disruption in Professional Services 

by Mark Meuldijk, Executive Management Consultant – AI & Data – Digital Assurance, PwC Switzerland

14h20

Talk: Disrupted timelines – How AI agents accelerate development and launch cycles

by Kevin O’Sullivan, Co-Director, ETHZ Agentic Systems Lab


14h45


Coffee Break

Part 4 – Governance

Address the complexities of governing, regulating, and assigning accountability within agentic systems, as both legal and ethical considerations are reshaped by the rise of autonomous AI agents. The discussion culminates with a panel exploring frameworks for effective oversight, responsibility, and redress.

15h15

Talk: Testing Autonomy: How Regulatory Sandboxes Can Support the Governance of AI Agents

by Clarissa Valli Büttow, FNS Senior Researcher, UNIL

15h40

Panel 2: Governance, accountability and regulation of AI agents

  • How should agentic systems be managed and monitored? What reporting is required? 
  • Who is responsible for the design, functioning, and decisions of these agents?
  • How to enforce governance and accountability? What happens in case of harm or failure?

Moderated by: [to be confirmed]

Panelists

  • Magdalena Barska, Senior Manager – Technology & Data Management, Accenture
  • Michel Jaccard, Founder, id est avocats
  • Petar Tsankov, CEO and Co-Founder, LatticeFlow AI
  • Clarissa Valli Büttow, FNS Senior Researcher, UNIL

Part 5 – Visionary

Conclude with a forward-looking dialogue on the future of agency and autonomy in AI, tackling profound questions of governance, safety, and control amidst the rise of self-improving, increasingly independent agents.

16h25

Visionary Panel: Agency, autonomy, and recursive self-improvement – governance and safety concerns

  • Autonomy versus Control
  • Paradigm Shift: The Disruptive Impact of Recursive Self-Improvement in AI Agents

Moderated by: [to be confirmed]

Panelists

  • Prof. Andrea Cavallaro, Head of L’IDIAP Laboratory, EPFL
  • Mark Meuldijk, Executive Management Consultant, PwC Switzerland
  • Mark Müller, Co-Founder and CTO, LogicStar AI
  • [to be confirmed]

17h10

Concluding Remarks

by David Viollier, C4DT

17h15

Networking Reception

Speakers

Nils Amiet
Lead Prototyping Engineer, Kudelski Security

Nils is a Security Researcher at Kudelski Security, specializing in AI security, fuzzing, privacy, and internet scanning. He has presented at Black Hat, DEFCON, and CCC, and authored fuzzomatic, an AI-powered Rust fuzzer. He has uncovered critical vulnerabilities in popular AI tools and helped build a distributed RSA key-breaking system. Nils also contributes to Kudelski’s research blog.

Magdalena Barska
Senior Manager – Technology & Data Management, Accenture

Julia Bazinska
Senior Research Engineer, Lakera AI

Julia is a Machine Learning Engineer at Lakera AI, developing their core product. Lakera AI is a frontrunner in AI security, known for empowering developers to build secure AI applications with Lakera Guard, which protects against prompt injections, data leaks, and other risks. Her career trajectory includes internships at Google, DeepMind, and IBM Research. She earned her bachelor’s degree in computer science from the University of Warsaw and completed her Master’s at ETH Zurich in 2023. Her professional interests are in Machine Learning for Natural Language Processing, AI security, and performance optimization of ML systems.

Andrea Cavallaro
Professor EPFL, Head of L’IDIAP Laboratory

Prof. Andrea Cavallaro is Professor at EPFL, where he leads innovative research in the fields of artificial intelligence, machine learning, and signal processing. Previously the Director of the Centre for Intelligent Sensing at Queen Mary University of London, he has a distinguished academic record and is recognized for his contributions to audio-visual processing, computer vision, and multimodal perception.

With a strong focus on developing technologies for robust and trustworthy intelligent systems, Prof. Cavallaro’s research bridges fundamental advances with real-world applications in areas such as surveillance, multimedia content analysis, and human-computer interaction. He is committed to advancing interdisciplinary collaborations and nurturing the next generation of researchers in the rapidly evolving fields of AI and signal processing.

Michel Jaccard
Founder, id est avocats

Michel Jaccard is the founder of id est avocats, a Swiss-based, independent law firm focusing on corporate and technology matters. For the past 25 years, he has been involved in corporate finance and M&A deals in Switzerland and abroad.

Michel provides counsel to companies on data protection, and cybersecurity. He resolves disputes in complex business projects involving advanced technologies and intellectual property licensing, including open-source and open-access strategies.

Michel holds a JD and PhD from the University of Lausanne and an LLM from Columbia Law School. Michel is admitted to practice in Switzerland and New York.

Martin Jaggi
Professor EPFL, Head of Machine Learning and Optimization Laboratory

Martin Jaggi is an Associate Professor at EPFL, heading the Machine Learning and Optimization Laboratory (MLO). Before that, he was a post-doctoral researcher at ETH Zurich, at the Simons Institute in Berkeley, and at École Polytechnique in Paris. He earned his PhD in Machine Learning and Optimization from ETH Zurich in 2011, and a MSc in Mathematics also from ETH Zurich.

Martin is a member of the Steering Committee of the Swiss AI Initiative and contributed to the development of Apertus, Switzerland’s fully open, transparent, multilingual language model.

Katherine Loh
Project Manager, Center for Digital Trust (C4DT), EPFL

Katherine Loh is a Project Manager for C4DT’s Agency team. She is helping C4DT build a global community of digital trust researchers and practitioners, through organizing events, connecting partners, and curating resources. Prior to joining EPFL, Katherine worked as a policy strategist in the international development domain, working with government, multilateral, industry and civil society organizations to articulate their responses to governance challenges posed by new technologies. Over time, she has witnessed the evolution of these challenges, from issues like mobile money deployment and small business digitalization to the future of work and mis- and disinformation. Katherine has a Masters in Economics from the University of California at Los Angeles.

Mark Meuldjik
Executive Management Consultant – AI & Data – Digital Assurance, PwC Switzerland

Mark Müller
Co-Founder and CTO, LogicStar AI

Dr. Mark Niklas Müller is a Co-Founder and CTO of LogicStar AI, where he leads the development of autonomous code agents for self-healing software systems. He pioneered methods for benchmarking software agent capabilities, including automated benchmark generation and the widely used SWT-Bench. Before founding LogicStar AI, he completed his PhD and Postdoctoral research at the Secure, Reliable, and Intelligent Systems (SRI) Lab at ETH Zürich under Prof. Martin Vechev, focusing on provable guarantees for machine learning systems and the intersection of AI and code.

Kevin O’Sullivan
Co-Director, ETHZ Agentic Systems Lab

Marcel Salathé
Professor EPFL, Co-Director, EPFL AI Center

Prof. Marcel Salathé is a renowned scientist at the intersection of health and computer science, currently serving as Professor and Co-Director of the EPFL AI Center. With a PhD in Biology and Environmental Sciences from ETH Zurich, he has held academic positions at Stanford University and Penn State before leading the Digital Epidemiology Lab at EPFL. Marcel is the founder of the EPFL Extension School, an initiative delivering high-quality online education in digital technology, and a strong advocate for lifelong learning in response to rapid technological advances.

An accomplished entrepreneur and innovator, Marcel founded AIcrowd.com, a leading AI challenge platform used by organizations such as OpenAI, Microsoft, and Google, and initiated the AMLD Intelligence Summit, one of Europe’s largest AI conferences.

Lea Strohm
Senior Technical Product Manager, Thomson Reuters

Lea has a multidisciplinary background in international relations and innovation studies, with a special interest in health innovation. She has conducted research on the implementation challenges of using AI in medicine, particularly in the field of medical imaging. Currently, Lea is a Senior Technical Product Manager on Thomson Reuters’ Enterprise AI Platform team, based in Zug, Switzerland, where she leads the design and delivery of enterprise-scale AI services to enable the creation of reliable, responsible, and trustworthy AI solutions. In this role, she focuses on developing and enhancing tools for AI evaluation to drive the adoption and continuous improvement of GenAI technologies across the organization.

Prior to joining Thomson Reuters, Lea worked at ethix – Lab for innovation ethics in Zurich, where she partnered with a variety of actors in the innovation ecosystem to assess ethical questions related to innovation in practical settings. Her multi-sector experience uniquely positions her at the intersection of technology, ethics, and policy, making her a key driver of responsible AI and innovation practices.

Petar Tsankov
CEO and Co-Founder, LatticeFlow AI

Petar is the co-founder and CEO of LatticeFlow AI. He holds a PhD from ETH Zurich, where he developed the first scalable frameworks for verifying the safety of deep neural networks, work that earned him the John Atanasoff Prize. Previously, he co-founded ChainSecurity, a leader in smart contract security, acquired by PwC. Under his leadership, LatticeFlow AI has received multiple awards, including the Swiss AI Award and recognition on the CB Insights AI100 list.

Clarissa Valli Büttow
FNS Senior Researcher, UNIL

Clarissa is a Senior Researcher and part-time lecturer at the Swiss Graduate School of Public Administration (IDHEAP), University of Lausanne. Her work explores the evolving relationship between public law and emerging technologies, with a particular focus on data governance, regulatory innovation, and the democratic implications of digital transformation. Her recent doctoral research, Public Sector Data Openness in the Crafting of the Data-driven Society, examined how regulation and innovation co-shape the development of data-driven societies. She continues to investigate how legal frameworks respond to technological change, especially in the context of AI governance and digital transformation.