5 AI Safety Sessions Every Founder Should Catch at TechCrunch Disrupt 2026

As artificial intelligence moves beyond demos and into real-world use — powering enterprise systems, autonomous vehicles, and physical robots — safety and security have become essential parts of the product itself, not afterthoughts. At TechCrunch Disrupt 2026, taking place October 13-15 at Moscone West in San Francisco, five sessions across the AI Stage and Real World AI Stage are tackling exactly these challenges. Here's a look at what founders can expect from each. 1. What Anthropic Sees When Enterprises Actually Deploy Claude Some companies see measurable results from AI deployment, while others remain stuck in pilot mode more than a year later. Anthropic Head of Applied AI Cat de Jong works directly with enterprises integrating Claude into critical workflows and has a front-row view of what separates successful deployments from stalled ones. In this AI Stage session, de Jong will explore the transition from experimentation to production, offering founders insight into why some AI implementations deliver real business value while others never move past the pilot phase. 2. The Agent Security Problem Nobody Is Talking About AI agents capable of taking autonomous action introduce a new category of security risk — what should an agent have access to, what actions should it be permitted to take, and what happens when standard application-level permissions fall short? Okta President of Products and Technology Ric Smith and NanoCo co-founder and CEO Gavriel Cohen will address agent security at the infrastructure level, examining weaknesses in current permission models and the architectural decisions founders need to weigh before deploying agentic AI systems. 3. Securing the AI Enterprise: Why the Cloud Just Got a Lot More Complicated Getting an AI product adopted by enterprise customers requires more than a strong demo — security, governance, and observability all become central concerns once AI takes on a role in critical business systems. AWS VP of Security Services Rudy Mitra, Luta Security CEO Katie Moussouris, and cybersecurity veteran Wendy Nather will discuss the infrastructure enterprises need as AI systems take on increasingly autonomous responsibilities, offering founders a look at what it takes to meet enterprise security expectations. 4. Building AI Systems When Failure Is Not an Option When AI operates in the physical world, the stakes of failure rise significantly — mistakes can affect vehicles, aircraft, industrial systems, and mission-critical operations rather than staying confined to a screen. Shield AI Chief Technology Officer Nathan Michael, General Motors Director of Robotics Strategy Mikell Taylor, and Waabi founder and CEO Raquel Urtasun will bring perspectives from defense and autonomous systems to address one of hard tech's toughest questions: how do you determine when an autonomous system is genuinely safe enough to deploy? The discussion will cover building a safety-first culture, testing and validation practices, and navigating regulatory hurdles. 5. Robots Are Waiting for Their ChatGPT Moment. Here Is What Is Standing in the Way Robotics faces a data problem — unlike language models and self-driving systems, robots lack access to the massive training datasets that fueled rapid progress in those fields. Nvidia Inception Global Head of Physical AI Les Karpas will examine how data pipelines, simulation environments, and foundation models could help close that gap, and what it will take to build robots capable of the kind of breakthrough moment that transformed language AI. What's at Stake for Founders These five sessions are part of a broader lineup of more than 200 sessions across six industry stages at Disrupt 2026, which is expected to draw more than 10,000 founders, investors, and operators alongside 250-plus speakers and 300-plus exhibiting startups. For AI builders, the underlying challenge each session addresses is the same: building a capable model, agent, or robot is only part of the task — earning enough trust from customers and enterprises to actually deploy it may be the harder problem to solve.