Innovation

Across industries, organizations are asking industrial systems to perform in conditions they were never designed for. The world has changed faster than the core logic of these systems, yet expectations continue to grow that AI will make them flexible, predictive, and reliable under uncertainty.

Data volumes continue to increase, but understanding does not keep pace. Tacit knowledge remains locked in people, operational data is fragmented across siloed systems, and many AI initiatives focus on optimizing isolated functions rather than preparing organizations to operate as coherent, adaptive wholes. This pattern repeats across industrial, defense, and space operations, where rare events, degraded conditions, and time-critical decisions are part of everyday reality.

As systems become more autonomous and AI-enabled, success depends not only on technical performance, but on whether humans are prepared to perceive what matters, learn from experience, and act coherently under real-world pressure. Without deliberate cognitive readiness, intelligence alone does not translate into resilience, trust, or safe operations.

This session positions AI within the broader concept of human cognitive readiness and traces its evolution—from static, procedure-based training to simulation-driven preparation, and toward AI-enabled, adaptive readiness systems, where preparation for even the most unexpected events takes place primarily in simulated environments rather than through costly real-world exposure.

Drawing on real-world work across industrial, defense, and space environments, Sini Rytky demonstrates how AI already improves cognitive readiness in practice, for example by:

Enhancing perception through multimodal data fusion and rich situational awareness

Supporting learning by identifying and understanding human behavioral patterns in complex operational contexts

Building on this foundation, the presentation outlines current market demands for readiness, including faster onboarding into complex roles, preparedness for rare and high-impact scenarios, and the ability to manage complex operations remotely.

Finally, the session introduces key research directions being explored within a leading research institution to address emerging readiness gaps, including:

AI-driven knowledge capture, enabling critical experiential expertise to be externalized and transferred

Synthetic data for AI model configuration and compatibility, supporting robust performance across evolving systems and environments

Distributed situational awareness, enabling shared, real-time understanding across humans, AI systems, and autonomous assets

Attendees will leave with:

Practical examples of how AI can contribute to cognitive readiness systems, and how readiness expectations are evolving across safety-critical domains

A clear, actionable checklist for initiating cognitive readiness planning in their operations, including key questions to assess the maturity of human preparedness, knowledge gaps, and simulation-based readiness systems

Building voice AI in Europe has never been easy. Strict regulations, a lack of EU-hosted solutions, and extremely high costs significantly increase the entry barrier—especially in industries such as FinTech or InsurTech.

However, difficult doesn’t mean impossible. In this talk, Artem Demchenkov presents a practical, cost-effective, and fully compliant EU-based approach to building high-quality voice AI without compromising performance.

Attendees will explore the three core components of voice AI: Speech-to-Text (STT), Text-to-Speech (TTS), and Large Language Models (LLMs). Artem will walk through how to configure and fine-tune each for improved performance, reduced latency, and more natural sound. He will also address key compliance considerations for each component, ensuring the solution aligns with EU regulatory standards.

Europe is at a crossroads: should it invest decisively in sovereign, responsible AI, or become increasingly dependent on U.S. Big Tech? The European Commission responded to this challenge on 8 October 2025 with a strong action plan, the “Apply AI Strategy” (see here), whose core objective is the development of sovereign generative AI models that comply with EU law and public values. If the concern is that Big Tech’s large language models (LLMs) do not align with European rules and societal values, Europe must innovate to change this reality.

The Dutch government rose to this challenge by awarding a grant to the Dutch National Institute for Applied Science (TNO) to develop a competitive, sovereign Dutch large language model fully compliant with EU legislation and public values: GPT‑NL. In this keynote, the speaker discusses the radically different approach taken—and the associated technical innovations required—to achieve this goal.

Early results show:

That high‑quality content enables the training of a high‑quality LLM with significantly less data, demonstrating that untargeted scraping of the internet is not a requirement.

That GPT‑NL outperforms comparable models in tasks such as text summarization, making it particularly suitable for trust‑critical government applications.

GPT‑NL offers a competitive alternative to current practices, where AI models are often trained on scraped internet data without the consent of data subjects or rights holders. Instead:

GPT‑NL does not scrape random data from the internet. It is the first model trained exclusively on texts that are either copyright‑free or covered by valid licenses. The Netherlands is also the first country worldwide to reach agreements with all publishers regarding content usage and fair remuneration—demonstrating that innovation and respect for intellectual property rights can coexist.

GPT‑NL is also the first model to apply data minimization at the source. TNO implemented an innovative contextual anonymization technique that removes all personal data of non‑public individuals while preserving sentence structure, enabling effective learning while ensuring material protection of affected individuals. These techniques also retain factual information about public figures, which legitimately belongs to the public domain

AI systems are rapidly evolving from passive chat assistants into autonomous agents capable of planning tasks, accessing sensitive data, invoking tools, and executing actions across enterprise environments. This shift fundamentally changes the security model. When AI systems move from generating suggestions to performing actions, they introduce new attack surfaces, privilege escalation paths, and governance challenges that traditional security controls were not designed to handle.

In this talk, the speaker examines how agentic AI systems operate from a security perspective and why existing access control, monitoring, and detection approaches break down in the presence of autonomous, probabilistic behavior. The session explores common failure modes such as overprivileged agents, prompt-driven misuse, tool abuse, and the lack of agent-level auditability. Rather than focusing on theoretical risks, the session frames agentic AI as an emerging execution layer that must be governed, monitored, and constrained like any other high-risk system component.

Attendees will leave with a clear understanding of how AI autonomy changes the threat model, along with practical guidance on the security principles required to safely deploy and operate autonomous AI systems in production.