Focus: Energy grids, global chip distribution, and the physical reality of compute capacity

While legislative and ethical frameworks dominate public discourse, the execution of artificial intelligence remains bound by physical realities. The sudden, explosive demand for advanced generative tools has forced a massive global restructuring of data centers, supply chains, and power distribution systems. This opening plenary shifts the focus to the foundational layer of the AI ecosystem, examining how regional energy grid limits, hardware access disparities, and severe talent shortages impact a nation’s true capacity for technological self-reliance. The panel will explore whether localised infrastructure autonomy is globally scalable, or if resource constraints are pushing the international community toward unavoidable cross-border tech dependencies.

• As capital investment into hyperscale data infrastructure hits record highs, what are the most viable strategies for balancing the immense energy and power requirements of front-line AI clusters with regional green transition mandates?
• With the rapid increase in localised data residency requirements across the EU and globally, how do we evaluate the trade-off between building sovereign regional compute centers and accessing centralised cloud networks?
• How can ecosystems with mature regulatory frameworks like the EU ensure they remain equally competitive in developing and scaling native physical compute capabilities and advanced hardware manufacturing?
• Given the highly concentrated global pool of advanced machine learning engineers, what practical frameworks can both established European tech hubs and emerging global markets implement to foster an equitable cross-border distribution of talent?

Focus: Harmonising cross-border standards across frontier labs, global AI Safety Institutes (AISIs), and digital safety commissioners

As advanced frontier systems migrate from theoretical research environments into critical public and commercial networks, the definition of AI safety must evolve from philosophical principles into uniform engineering practices. State-backed AI Safety Institutes, frontier research labs, and independent regulators are tasked with auditing complex models, tracking multi-turn conversational risks, and establishing default protection mechanisms. This session investigates the operational challenges of executing technical risk management across separate jurisdictions. Panelists will look closely at how the global technical community can align red-teaming, open-ended evaluation loops, and user protection protocols to prevent international security blind spots while reducing duplicate compliance burdens for software developers.

• When an advanced risk evaluation reveals a systemic vulnerability or open-ended safety flaw in an active commercial model, what protocols should guide the timeline between state public-safety notifications and corporate patch management?
• As open-weight model architectures approach parity with proprietary ecosystems, what frameworks effectively balance the economic value of open-source innovation with the need to prevent malicious modifications of safety controls?
• In light of increasing focus on open-ended generative interactions, what technical frameworks and early-detection parameters can frontier labs implement to monitor safety during multi-turn conversational models?
• What are the primary operational steps required to reconcile EU and Asian safety baselines with US/UK testing frameworks to reduce duplicate compliance friction for multinational developers?

Focus: Cultivating regional language models while maintaining global interoperability

The current concentration of AI foundational research poses a complex challenge for nations seeking to preserve their cultural, linguistic, and economic sovereignty. Standard models trained predominantly on Western-centric, English-dominated internet corpora often fail to capture the nuances or meet the practical needs of emerging markets and regional communities. This session focuses on the real-world trade-offs of building native-language LLMs and state-funded tech stacks from scratch versus fine-tuning global open-source platforms. The discussion will address how to foster inclusive economic development, reduce digital divides, and build equitable data ecosystems without resorting to isolationist data policies.

• For markets looking to establish digital self-reliance, what are the key factors in deciding whether to invest in building localised foundation models from scratch versus fine-tuning existing open-source frameworks?
• Given that many global foundation models are trained on English-dominated datasets, what technical and policy mechanisms can best integrate European regional languages and Global South contexts into mainstream AI architectures?
• How can the global AI ecosystem shift the role of emerging economies from predominantly providing downstream data-labeling labor to actively participating in core upstream IP development and revenue generation?
• How can policymakers structure data-localisation strategies so they protect citizen privacy and domestic interests without unintentionally separating local startups from international research partnerships?

Focus: Enterprise case studies, legacy stack migration, real-time platform safety, and scaling ROI.

The initial phase of generative AI experimentation has passed, leaving global enterprises and interactive platform operators with the far more difficult task of scaling tools into core production environments. Transitioning from temporary proof-of-concept projects to durable, daily automation requires organisations to overhaul legacy IT infrastructures, deploy real-time automated moderation systems, manage complex data privacy risks under GDPR, and redefine human-machine workflows. This panel centers on the operational realities of AI diffusion within highly regulated environments. Participants will share evidence-based insights into calculating actual return on investment (ROI), assigning systemic liability when autonomous agents fail, and maintaining workplace trust during large-scale operational shifts.

• As the industry transitions from trial phases to long-term integration, what are the primary structural bottlenecks preventing companies from scaling autonomous AI agents into live production environments?
• When autonomous enterprise or interactive conversational systems execute flawed transactions or unexpected outputs, how should liability be cleanly partitioned between the foundation model developer, the third-party integrator, and the deployer?
• How do enterprise platform operators manage the technical constraints of deploying real-time automated safety filters and moderation models at global scale without causing server latency or breaking privacy rules?
• What metrics and data-minimisation practices have proven most effective at building internal employee and user trust during large-scale workflow automation initiatives?

Focus: Turning high-level policy into technical engineering benchmarks

High-level political declarations and legislative text establish the boundary lines for AI development, but technical standards are what translate law into code. Standard-setting bodies like CEN-CENELEC, ISO, and the ITU are now under immense pressure to design product engineering benchmarks that match the rapid pace of machine learning innovation. This session looks at the practical mechanics of global interoperability, evaluating how technical specifications can bridge diverging regulatory definitions across regions. Panelists will debate how to establish reliable, international hardware and software verification baselines, such as digital watermarking, to protect data and information integrity globally.

• Given that machine learning research is moving faster than administrative timelines, how can international and European standard bodies create engineering benchmarks that remain relevant by the time they are published?
• What are the most practical pathways for harmonising core technical definitions, such as what constitutes “systemic risk” or “high-risk deployment”, between the EU and major international markets?
• How can standard-setting bodies ensure that the technical engineering specifications required by new legislative frameworks are finalised quickly enough to give businesses a clear roadmap to compliance?
• Can hardware and software-level tracking systems, such as digital watermarking, reliably secure cross-border information integrity, or do open-weight ecosystems make standardising content attribution technically unfeasible?

Focus: Executive regulatory compliance, EU AI Office enforcement, and cross-border trade impacts

As the enforcement of major regulatory frameworks like the EU AI Act takes effect, its structural requirements are reshaping the international tech marketplace. For organisations operating across borders, navigating steep non-compliance penalties and localised enforcement mechanisms introduces a critical layer of strategic risk. This flagship closing session brings together top European Commission leadership, international state leads, and corporate executives to analyse how regional compliance mandates interact with agile commercial deployment and foreign market entry rules. The panel will discuss how to prevent global market fragmentation and ensure that strict safety guardrails support, rather than restrict, competitive technological transformation.

Opening Keynote Address (16:55 – 17:10)

Panel Discussion (17:10 – 18:00)

• As significant financial penalties under the EU AI Act introduce new risk dynamics, how can regional regulators in Brussels ensure compliance enforcement remains predictable and supports market entry for international firms?
• With individual nations enacting their own rules around data governance and online safety, what diplomatic tools are most effective at preventing international supply chain fragmentation?
• Do current commercial developments, such as the creation of European sovereign clouds, represent a long-term shift toward decentralised global infrastructure, or are they a temporary reaction to localised compliance requirements?
• For regions undergoing rapid digital transformation, how can policy frameworks balance the introduction of robust safety gates with the agile speed required to support national modernisation and economic resilience?